Algorithm 916: computing the Faddeyeva and Voigt functions

Size: px
Start display at page:

Download "Algorithm 916: computing the Faddeyeva and Voigt functions"

Transcription

1 Algorithm 96: computing th Faddyva and Voigt functions MOFREH R. ZAGHLOUL, Unitd Arab Emirats Univrsity AHMED N. ALI, Unitd Arab Emirats Univrsity W prsnt a MATLAB function for th numrical valuation of th Faddyva function w(z). Th function is basd on a nwly dvlopd accurat algorithm. In addition to its highr accuracy, th softwar provids a flxibl accuracy vs fficincy trad-off through a controlling paramtr that may b usd to rduc accuracy and computational tim and vic vrsa. Vrification of th flxibility, rliability and suprior accuracy of th algorithm is providd through comparison with standard algorithms availabl in othr libraris and softwar packags. Catgoris and Subjct Dscriptors: G..0 [Numrical Analysis]: Gnral-Computr Arithmtic; Numrical Algorithms; Multipl prcision arithmtic; G.. [Numrical Analysis]: Approximation-Spcial Functions Approximations; G.4 [Mathmatical Softwar]: Algorithm dsign and analysis; fficincy Gnral Trms: Algorithms Additional Ky Words and Phrass: Function valuation, accuracy, Matlab, Faddyva function Author s addrsss: M. Zaghloul and A. Ali, Dpartmnt of Physics, Collg of Scincs, Unitd Arab Emirats Univrsity, Al-Ain, 755, UAE.

2 . INTRODUCTION Th valu and significanc of th scald complmntary rror function for complx variabls, also known as th Faddyva function or th plasma disprsion function, is wll rcognizd in th litratur for its applications in svral filds of physics [Armstrong 967; Gautschi 967; 970; Hui t al. 978; Humlíčk 98; Dominguz t al. 987; Popp t al. 990; Lthr t al. 99; Schrir 99; Shippony t al. 993; Widman 994; Wlls 999; Luqu t al. 005; Ltchworth t al. 007; Abrarov t al. 00]. Plasma spctroscopy, nuclar physics, radiativ hat transfr, and nuclar magntic rsonanc ar a fw xampls of th filds for which fficint and accurat valuation of this function is rquird. Som of ths applications rquir a small numbr of valuations of th function whr accuracy is mor important than th computational tim whil othr applications rquir normous numbrs of function valuations, which imposs tight rstrictions on th computational tim. Accordingly, computational accuracy and computational tim ar issus of intrst that should b critically addrssd and invstigatd in dvloping any succssful algorithm for th computation of this function. Motivatd by its practical importanc and a lack of closd form xprssions for th calculation of th Faddyva function, numrical valuation of th function has bn th focus of rsarch ovr many dcads [Armstrong 967; Gautschi 967; 970; Hui t al. 978; Humlíčk 98; Dominguz t al. 987; Popp t al. 990; Lthr t al. 99; Schrir 99; Shippony t al. 993; Widman 994; Wlls 999; Luqu t al. 005; Ltchworth t al. 007; Abrarov t al. 00; Zaghloul 007]. As a rsult, a wid varity of algorithms for th calculation of this function hav bn dvlopd and prsntd in th litratur. Howvr, as it is shown in this study, most of ths algorithms los accuracy in som rgions of th computational domain. W introduc a nw algorithm for th calculation of th Faddyva function which provids flxibility, rliability and suprior accuracy. In sction w prsnt th dfinition of th function and brifly summariz som rlvant fundamntal mathmatical rlations. Thn, in sction 3, w stablish th analytical basis of th algorithm whil numrical analysis and computational dtails ar discussd in sction 4. A short dscription of th Matlab function is givn in sction 5. Vrification of th algorithm and comparisons with othr comptitiv cods in th litratur ar providd in sction 6.. DEFINITION AND FUNDAMENTAL MATHEMATICAL RELATIONS For a complx variabl z=x+iy, th Faddyva (plasma disprsion) function, w(z), th ral Voigt function, V(x,y), th imaginary Voigt function, L(x,y), th complx rror function, rf(z), th imaginary rror function, rfi(z) and th Dawson s intgral F(z) ar all closly rlatd to ach othr. On can summariz ths rlations as w( z) = = = z z z ( rf( iz)) = (+ rf( iz)) (+ i rfi( z)) z i = + F( z) = V ( x, y) + i L( x, y) z rfc( iz) for y > 0 ()

3 In th abov rlations, i =, rfc(z) is th complmntary rror function, rfi(z) is th imaginary rror function which is rlatd to th rror function by rfi(z)= - i rf(iz). As can b sn from th last lin in (), th ral and imaginary Voigt functions ar just th ral and imaginary parts of th Faddyva function for y>0, rspctivly. Th valuation of all of ths functions can, thrfor, b prformd through th rror function. Th rror function of a complx variabl z can b rgardd as a lin intgral in th complx plan givn by z = t rf ( z) dt () 0 Diffrnt paths can b takn to prform this lin intgral in th complx plan. For xampl, on may choos a linar path btwn th initial point (origin) and th final point z which givs an xprssion for th complx rror function of th form z = z t rf ( z) dt (3) 0 An altrnativ path can b followd through th lin sgmnts [0, 0 iy] and [0 x, iy] which givs, for th complx rror function th xprssion, x y y y x = t t + t rf ( z) cos(yt) dt i dt sin(yt) dt (4) In addition to th abov paths, anothr simpl and usful path from th initial to th final points can b takn through th lin sgmnts [0 x, iy=0] and [x, 0 iy] which rsults in x y y t t t rf ( z) = dt + sin(xt) dt + i cos(xt) dt (5) Th first trm on th right hand sid of (5) is th dfinition of th rror function of th ral variabl x. Equation (5) is th basis of th prsnt algorithm for th calculation of th Faddyva function as shown blow. For som limiting valus of th paramtrs x and y, analytical formula do xist for th ral and imaginary parts of th Faddyva function V ( x 0, y) rfcx( y) L( x 0, y) 0 V L ( x + y ) ) ( )[ y ( x + y )] ( x + y ) ) ( )[ x ( x + y )] V ( ± x, y 0) (6) y whr rfcx( y) = rfc( y) is th scald complmntary rror function of th ral argumnt y. Whn y 0, th imaginary part of th Faddyva function cannot b xprssd as simply, howvr, it can b xprssd in trms of Dawson s intgral of x (th ral part of z) whr L( ± x, y 0) F( x) (7) 3

4 and F(x) has wll rportd asymptotic xprssions for limiting valus of x [Armstrong 967]. Following Salzr 95 [Salzr 95], w can writ t a ( + a n t cosh (ant) ) ± E a n= whr th rlativ rror n / a E = cos(n t / a) is of th ordr of / a E ~. n= Tabl prsnts som rprsntativ valus of th rlativ rror E corrsponding to som valus of th paramtr a. Tabl : Rlativ rror, E, of th rprsntation of paramtr a. t givn by Eq. (8) as a function of th a E (8) 3- ANALYTICAL BASIS OF THE ALGORITHM t Rplacing in (5) by its rprsntation givn in (8), w gt, for th ral and imaginary parts of rf(z), th following xprssions R[rf( z)] = rf( x) + n= a x a n 4a n + 4 x 8a ( cos(xy)) + [ x x cosh(a n y) cos( x y) + a n sinh(a n y) sin( x y) ] (9) 4a Im[rf( z)] = n= (4x) a n 4a n + 4 x 8a sin( x y) + [ x cosh(a n y) sin( x y) + a n sinh(a n y) cos( x y) ] (0) Th rror in th xprssions in (9) and (0) for th ral and imaginary parts of th complx rror function is controllabl through th paramtr a as indicatd in Tabl. Both xprssions, howvr, rduc to th formula givn by Salzr [Salzr 95] and Abramowitz and Stgun [Abramowitz t al. 97] with a rlativ rror lss than th floating point rlativ accuracy, ε, on a 6 digit computational platform, by stting a qual to /. An xprssion for th Faddyva z function, w ( z) = R[ w( z)] + i Im[ w( z)] = rfc( iz), can b obtaind by substituting (9) and (0) into (). Th rsulting xprssions for th ral and imaginary parts of th Faddyva function ar thn givn by 4

5 R[ w( z)] = 8a y + Im[ w( z)] = 8a + a xsin( xy) rfcx ( y) cos(xy) + n= 4 a n 4 a n a n + 4 y a x rfcx ( y) sin (xy) + n= a n + 4 y [sin( xy) ( xy)] [cosh(a n x) cos( x y)] [sin(xy) (xy)] [ y sin(xy) + a n sinh(a n x)] () () Som commnts on th abov xprssions may b usful at this stag - Th sris in () and () hav an infinit numbr of trms and nd to b truncatd for practical us. Th ffct of such a truncation on th accuracy of th computations is prdictabl and can b controlld for a convrging sris, - Whil th xponntial factors in th sris ar dcaying with th indx, n, thy ar multiplid by growing hyprbolic functions and on cannot dirctly dtrmin whr to truncat th infinit sums for practical us, 3- Thr is a limitation on th valuation of th hyprbolic functions (rlatd to th largst positiv floating point numbr, R max, availabl on th computational platform). Sinc th argumnt of ths hyprbolic functions is (anx), this will impos a strict rstriction on th numbr of trms to b includd in th sums for a givn valu of x with possibl catastrophic consquncs on both accuracy and rliability, 4- In writing th abov xprssions, th scald complmntary rror function of a ral variabl is usd to rduc rounding rror associatd with th trm (-rf(y)), howvr, spcial car is ndd for th valuation of th rfcx(y) function to avoid ovrflow problms illustratd in [Zaghloul 007]. At prsnt, many softwar packags hav wll-bhavd algorithms for computing th scald complmntary rror function of a ral variabl, rfcx, and algorithms for accurat and fficint computation of this function ar availabl in th litratur, 5- Th valuation of th quantity xp(-x ) common to all of th abov trms, can suffr undrflow problms for larg valus of x. Ths problms can b avoidd by combining this quantity with othr larg quantitis whrvr possibl. To ovrcom th abov-statd concrns and rstrictions, th xprssions in () and () ar rwrittn in th forms sin( ) x a x xy R[ w( z)] = rfcx( y) cos(xy) + [sin( xy) ( xy)] (3) a y y + y cos(xy) Σ + Σ + Σ 3 and 5

6 x a x Im[ w( z)] = rfcx( y)sin(xy) + [sin(xy) /(xy)] (4) a + y sin(xy) Σ Σ 4 + Σ 5 whr ( a n + x ) n= a n y Σ = (5) + ( an+ x) n= a n y Σ = (6) + ( a n) 3 n= a n y Σ = (7) + an + ( a n x) Σ 4 = (8) n= a n + y a n ( an) Σ 5 = (9) n= a n + y Th convrgnc of th sris (5)-(9) can b vrifid by applying th simpl ratio-tst [Boas 006]. In addition, in all of th abov xprssions (5)-(9), th pr-xponntial factors (fractions in brackts) of th argumnts of th summations assum valus lss than for n>/a. For a=/ (which is sufficint to xprss by th xprssion givn in (8) to machin accuracy on a 6-digit computational platform, s Tabl ) th prxponntial factors will always assum valus lss than or qual to on for n. For valus of n, th trms in th sris Σ, Σ and Σ 4 dcras monotonically with n whil for rlativly larg valus of x th trms Σ 3 and Σ 5 incras with n from on to a crtain limit whn thy start dcaying monotonically with n. Th fact that all xponntial factors in th abov summations vntually dcay with incrasing n provids us with th possibility of obtaining a truncatd sris of practical us and computational fficincy as shown in th nxt sction. It is undrstood, howvr, that all of th abov summations ar to b prformd using a singl loop for computational fficincy. Th loop spcifications can b dtrmind onc a cutoff schm for th sris in (5)-(9) is stablishd. Morovr th computation of th xponntials in ths sris can b rducd to th computation of a singl xponntial and products using a singl computational loop. This rducs th dpndnc on using th intrinsic function to calculat ths xponntials within th loop and can sav a significant amount of computational tim. As th ral part of th Faddyva function is vn in x and its imaginary part is odd in x, w nd only considr th right half of th complx plan (x 0) sinc th vn/odd proprtis of th ral and imaginary parts of th function can b usd to find th corrsponding valus in th lft half of th plan. In addition, th valus of w (z) in th lowr half of th complx plan can b obtaind from valus in th uppr half using th rlationship z w( z) = w( z) t (0) 6

7 which is quivalnt to th symmtry rlations V ( x, y) = L( x, y) = x + y + y cos(xy) V (, y) sin(xy) L(, y) (0 ) Thus w do not los gnrality by considring th valuation of th function in just th first quadrant, howvr, w not that (0) rquirs th subtraction of two valus which lads to a loss of accuracy in th lowr half. Th practical usfulnss of th calculation of th partial drivativs of th ral and imaginary parts of th Faddyva function has bn pointd out by many authors [0,3,5]. Onc th Faddyva function has bn calculatd accuratly on can calculat ths partial drivativs with rlativ simplicity using th xprssions V ( x, y) = R[ zw ( z)] = [ y L( x, y) xv ( x, y)] () x V ( x, y) = Im[ zw ( z)] = [ x L( x, y) + yv ( x, y)] () y in conjunction with th rlations L( x, y) V ( x, y) L( x, y) V ( x, y = & = ) (3) y x x y Schrir [Schrir 99] rports numrical problms that can aris whn subtracting two numbrs of approximatly qual magnituds (whn V x ~ 0 ). Ltchworth and Bnnr [Ltchworth t al. 007] us a spcial algorithm to calculat ths partial drivativ (to accuracy <0.5%) but this incrass th computational tim by about 70%. 4- NUMERICAL ANALYSIS AND MACHINE LIMITATIONS 4.. High Accuracy Computations Du to th finit numbr of dcimal digits availabl to stor a ral numbr in floating-point arithmtic, thr ar machin limitations on th valuation of th abov summations. Th floating-point rlativ accuracy, ε, and th smallst positiv floatingpoint numbr, R min, in th usd computational platform impos rstrictions on th accuracy and caus practical machin-truncation of th sums. At any stag during th computation of th sums, th nw accumulatd sum aftr adding th trm α n+ of th sris can b writtn as Σ = Σ + α ( ( n) ) n n n+ = Σ +, (4) + n whr ( n) = α n+ Σ. n (4) implis that th sum of n+ trms will not diffr from th sum of n trms (i.. th sris will b ffctivly machin-truncatd) if th trm ( n) = α n+ Σ bcoms n lss than th floating-point rlativ accuracy, ε, or if α n+ <R min. For computational fficincy (shortr computational tim) w nd to spcify ths intrnally truncatd trms and xclud thm from th computational loop. Starting with th sum Σ and considring th possibility of machin-truncation of th sum du to th undrflow of th trms α n+, a simpl saf stimation for th last valu 7

8 Σ of th indx n to b includd in th valuation of th sum, n cut, can b drivd basd on th undrflow of th xponntial factors only (sinc th pr-xponntial factor is alrady lss than unity) Σ n ln( R ) x cut, R min (5) min a It is implicitly undrstood that (5) implis rounding to th narst intgr towards infinity. Similarly, th trms of th sums, Σ and Σ 4, hav th sam xponntial dpndnc, whil both of th pr-xponntial factors will hav valus lss than or qual to unity for n / a. In such a cas, th trm α n+ in both sums will rach valus <R min if th xponntial factor bcoms R min which givs Σ, Σ 4 ncut, R [ ln( Rmin) x] (6) min a Both (5) and (6) indicat that, for x ln( R min ), no trms from th argumnts of th sums Σ, Σ and Σ 4 will b ffctiv in th computations and that th valus of ths sums will b ffctivly truncatd. W not, howvr, that machintruncation of ths sums du to th R min limitation would also imply machin-truncation du to machin accuracy (i., (n) ε ) as sn in (4). For computational fficincy, this latr condition may b usd to brak th computational loop as additional cycls of th computations or additional trms of th sris will not chang th valus of th sums. Furthrmor, w can also mak th computations of th sums Σ 3 and Σ 5 vry fficint. As pointd out abov, th valus of th trms of ths two sris grow initially with n up to a crtain valu (pak) and thn dcay continuously as n incrass. Simpl invstigation of ths two sums shows that this pak is in th vicinity of n=x/a. Accordingly, if on starts calculating ths sums from around n=x/a and procds in both dirctions, th valus of th trms will dcras until thy gt machin-truncatd. For ach valu of th indx n usd in th calculation of th trms of th sums Σ, Σ and Σ 4 w can add to ach of th sums Σ 3 and Σ 5 two trms by marching on stp in ach dirction. Th sums ar truncatd whn th sum of th nwly addd two trms rlativ to th valu of th prviously accumulatd sum bcoms lss than th machin accuracy. Handling th computation of th sums, Σ 3 and Σ 5 this way lads to a significant saving in xcution tim by dramatically rducing th numbr of trms rquiring valuation which, in turn, lads to a smallr numbr of loop cycls. Th asymptotic xprssion in th first lin of (6) is usd for valus of x<r min. 4.. Accuracy Vs Efficincy Trad-off With 6-digit floating-point arithmtic and for valus of th paramtr a > ½, th xpansion in (8) bcoms lss accurat and its accuracy will b govrnd by th corrsponding valu of th rlativ rror, E, as shown in Tabl. It is not ncssary, thrfor, to kp th strict condition for truncating th sums, i.., (n)<ε sinc th accuracy of th computations will b govrnd principally by th rlativ rror E. Rcalling that th rlation btwn a and E can b writtn as E ~ /a it sms mor appropriat to choos E (namd tiny in th cod) instad of ε to tst for th convrgnc of th sums. That way w can rduc th numbr of trms includd in th sums by xcluding trms that will not ffctivly nhanc th accuracy and thus achiv 8

9 rasonabl acclration of th computations and rduc th computational tim. Noting that, changing th valu of a changs E (tiny) and vic vrsa, w could choos ithr a or tiny as th fr paramtr for controlling th accuracy and fficincy of th computations. This fr paramtr will b usd as an argumnt of th function. W hav chosn to us tiny and w calculat a intrnally from th abov rlation as tiny givs a bttr indication of th accuracy of th computations. This allows flxibility for accuracy vs fficincy trad-offs whil maintaining th ability to run th cod for high accuracy Calculation of th Exponntials and Othr Numrical Considrations Th cntral part of th prsnt algorithm dpnds on th valuation of th sums (5)-(9) which all hav xponntial trms. Th valuation of intrinsic functions lik th xponntial function is known to b slowr than othr simpl mathmatical oprations such as multiplication and/or division. A naïv computation of ths sums would rquir thr xponntial valuations pr computational loop. This would b computationally xpnsiv. Howvr, w may rduc this to just on xponntial valuation in ach cycl of th loop. For th cas of x < R ) w hav som flxibility in valuating th x ln( min trm ithr sparatly or by combining it with othr trms. In such a cas on can a writ all of th xponntials in (5)-(9) in trms of th xponntial n whr ( a n + x ) a n x = (7) + n ( a n x) a n x = ax n ( an) a n ax = x ax In th abov xprssions ax, and ar calculatd onc outsid th loop, for ach valu of x, and th products ar simply prformd using multiplis insid th computational loop. For x R ) only Σ 3 and Σ 5 (which hav th sam ln( min xponntial factor) contribut to th calculation of th ral and imaginary parts of th Faddyva function. Howvr, du to th natur of th computation of ths trms, th xponntial factor nds to b computd twic; onc for th stp to th right of n 0 =cil(x/a) and onc on th lft wing whr cil indicats rounding to th narst intgr towards infinity. Th indics for ths two factors ar rlatd to th loop indx, n, by n 3- plus = n 0 +(n-) and n 3-minus = n 0 - n if n 3-minus, rspctivly. Whn n 3-minus <, only th trm on th right wing is includd. Th xponntial factors for th trms on th lft and right wings ar rlatd by n a n x ( an3 plus x) ( 3 minus ) ( a + ax a n0 ) (4a n0 4ax a ) = (30) Clarly th scond xponntial factor on th right hand sid of (30) and th argumnt of th product may b calculatd onc outsid th loop, for ach valu of x, thus rducing th numbr of xponntial function valuations to only on pr cycl. Th product can b valuatd using just multiplis insid th loop, thus rducing th computational tim. (8) (9) 9

10 A fw mor important computational points ar rlatd to th calculation of th imaginary part of th Faddyva function using (4). Firstly, for valus of x<<, th two sums Σ 4 and Σ 5 bcom vry clos to ach othr and th subtraction of ths two sums could significantly affct th accuracy in rgions of th computational domain whr ths two trms ar th dominant trms in calculating th imaginary part of th Faddyva function. Howvr, this problm can b simply ovrcom by xprssing th sum of ths two trms (for x<<) in its original form as in (), i.. in trms of sinh(anx). Th first thr trms in th sris xpansion of sinh(x) will b sufficint to xprss sinh(x) to th machin accuracy for x 0 - and sinc an is usually <0 in th prsnt computations thn this is satisfactorily for x Th scond important point in th calculation of th imaginary part of th Faddyva function using (4) is rlatd to th calculation of th sum of th first thr trms on th right hand sid of th quation, for x < ln( R min ), which can b writtn safly in th form + + a n a sin ( xy) rfcx ( y) (3) y n= + a n y Th trms in th curly brackts ar only dpndnt on y and for y 5 w hav found that this sum is zro to machin accuracy. Using this prvnts rounding rrors affcting th accuracy of computations. Not that for vry small valus of x th rsult of th whol xprssion (3) is O(yx) whil th total of -Σ 4 + Σ 5 is O(x), th significanc of rounding rrors is thus clar for small valus of x and rlativly larg valus of y. 5. THE MATLAB FUNCTION FADDEYEVA.M Th function Faddyva(z,tiny) rturns, in gnral, an array of complx valus for th Faddyva function of th sam siz as th input array for th complx variabl z. Th input z is usually an array (with on or two dimnsions) but can b a singl scalar as wll. Whn z contains only imaginary valus z=iy, th function rturns th ral valus calculatd from th MATLAB built-in function rfcx(y). Th function is st for th calculation for th whol complx domain. Howvr, for ngativ valus of y and xp(y - x ) gratr than th largst floating point numbr in th computational platform, Faddyva cannot calculat th Faddyva function du to inscapabl ovrflow problms. Th function chcks for accptabl valus and issus an rror mssag for any points outsid this domain. Th valu of th scalar fr paramtr tiny can b chosn by th usr within th rang tiny min tiny 0-4 to control th accuracy and computational tim. Th valu of tiny min is a valu clos to but lss than th floating-point rlativ accuracy, ε. For xampl, for a 6 digit computational platform, tiny min can b takn roughly to b ~ (th valu of E corrsponding to a=/ in Tabl ) whil for a 3-digit computational platform tiny min can b takn to b roughly Th maximum valu of tiny=0-4 corrsponds to a= (th maximum valu for a for which th xpansion in (8) can b usd). Incrasing th valu of tiny within its abov mntiond rang will dcras th computational tim at th xpns of th computational accuracy and vic vrsa. Choosing a valu of tiny<tiny min will just incras th run tim without any improvmnt in th accuracy of computations which will thn b govrnd solly by th 0

11 machin charactristics. Valus of tiny<tiny min or tiny>0-4 rsult in tiny bing rst to tiny min or 0-4 rspctivly; a warning mssag is rturnd in both cass. It is to b notd that tiny<ε is usd only for th calculation of th corrsponding valu of th paramtr a and not for th truncation of th sums sinc th calculations cannot b claimd to b prformd for rlativ accuracy lss than th machin accuracy psilon, ε, in any cas. Accordingly, for th truncation of th sums, th maximum of tiny and ε is usd. 6. ALGORITHM VERIFICATION AND EFFICIENCY 6.. High Accuracy Computations Thr diffrnt indpndnt computational tchniqus ar usd to invstigat th accuracy of th prsnt algorithm - Mathmatica [Wolfram Rsarch, Inc. 008] provids th imaginary rror function rfi(z) as a spcial function which can b valuatd and thn usd in conjunction z with th rlation givn in (); that is w ( z) = ( + i rfi( z) ), to calculat th Faddyva function [Zaghloul 008]. Th arbitrary-prcision arithmtic usd in Mathmatica allows us to obtain highly accurat valus for th function rfi(z) although ths calculations ar vry xpnsiv computationally. This is only gnrally suitabl for applications whr th spd of arithmtic is not a rstrictiv factor, or whr prcis rsults for a small numbr of valuations ar rquird. W can, howvr, gnrat highly accurat valus of th function rfi(z) using Mathmatica by using larg numbrs of digits of prcision, - th simpl propr intgral givn in rfrnc [Zaghloul 007] can b usd to calculat th ral Voigt function (ral part of th Faddyva function), and 3- th Algorithm 680 [Popp t al. 990], which is widly usd in th litratur and is implmntd in many softwar packags and libraris, calculats th Faddyva function to a claimd accuracy of 4 significant digits. Th rlativly long computational tim associatd with th first two mthods maks thm infficint for us in applications rquiring a larg numbr of function valuations. For this rason, Algorithm 680 is rgardd as th comptitiv highly accurat algorithm du to its computational spd and claimd accuracy. Tabl prsnts sampl rsults calculatd using ths thr mthods with th rsults from th prsnt algorithm. Th valus of th complx variabl z usd in th computations in this tabl hav bn slctd to allow som conclusions to b drawn in addition to stablishing confidnc and rliability in th prsnt cod. Th valu of th rlativ rror in th calculation of th Faddyva function proposd hr is givn in Tabl 3 and compard with th othr approachs, taking th valus of w calculatd using th function rfi(z) from Mathmatica with high numbr of digits of accuracy as rfrnc valus. Looking closr into th valus givn in ths two tabls w conclud that; a) compard to calculating th Faddyva function using rfi(z) from Mathmatica, th prsnt algorithm is mor rliabl sinc w faild to calculat rfi(z) using Mathmatica for valus of x> and y> whil th

12 prsnt algorithm dos not suffr such a limitation. Not that for this domain Mathmatica cannot b usd as rfrnc, and thrfor, no comparison was prformd for this rang in Tabl 3. b) for th whol domain of computations, th prsnt algorithm shows vry high accuracy as shown in Tabl 3, whil th Algorithm 680 suffrs a catastrophic loss of accuracy in th vicinity of x=6.3 as wll as for small valus of y signifid by bold-fac numbrs in Tabls and 3. Th rlativ rror for th ral part of th Faddyva function from Algorithm 680 in this rgion of th first quadrant gos up to 00%. It has to b notd that x=6.3 is on of th built-in valus in Algorithm 680. To invstigat th ffctivnss of th prsnt algorithm compard to Algorithm 680, w calculatd th Faddyva function using both algorithms for,840,70 points of th complx variabl z distributd ovr th uppr half of th complx plan using th grid y=logspac(-0, 4, 7) and x=linspac(-00, 00, 4000) whr y=logspac(-0, 4, 7) gnrats a row vctor of 7 logarithmically qually spacd points btwn 0-0 and 0 4 and x=linspac(-00, 00, 4000) gnrats a row vctor of 4000 linarly qually spacd points btwn -00 and 00. Using Matlab (R009b), th computational tim takn by th prsnt algorithm was found to b <8.0% of that takn by Algorithm 680, which rprsnts a significant tim saving. Figurs (-a) and (-b) show surfac plots of th absolut rlativ rror δ V = V Vrf / Vrf and δ L = L Lrf / Lrf in th rsults obtaind from th prsnt algorithm using rsults from Algorithm 680, which is availabl in Matlab, as rfrnc valus. As can b clarly sn from ths figurs, th rsults from th prsnt algorithm show high agrmnt (around 3 significant digits) ovr th chosn computational domain xcpt for th rgion in th vicinity of x=6.3 and small valus of y whr Algorithm 680 badly loss its accuracy as indicatd abov. Th abov findings provid th ncssary vrification and confirm th high accuracy as wll as th rliability of th prsnt algorithm.

13 Figur -a Absolut rlativ rror δ V = V Vrf / Vrf in th calculations of th ral part of th Faddyva function from th prsnt algorithm using th rsults from Algorithm 680 as rfrnc valus. Figur -b Absolut rlativ rror δ = L L / L in th calculations of th imaginary part of th L rf Faddyva function from th prsnt algorithm using th rsults from Algorithm 680 as rfrnc valus. rf 3

14 6.. Efficint Computations with Lowr Accuracis Tabl 4 blow shows valus of th fr variabl tiny usd in th calling argumnt of our Matlab function and th corrsponding rlativ accuracy δ = ( V V )/ V and δ L = ( L Lrf )/ Lrf in th calculations, using valus calculatd with th highst accuracy obtainabl from th prsnt algorithm as rfrnc valus. Th nd to quantify th fficincy improvmnts obtainabl whn using th accuracy vs fficincy trad-off capability of th prsnt algorithm is th rason of using th highst accuracy computations from th prsnt algorithm as rfrnc valus. Th run tims rquird to calculat th function for,840,70 points gnratd using th grid y=logspac(-0, 4, 7) and x=linspac(-00, 00, 4000) rlativ to th run tim rquird to prform th sam computations using th highst accuracy computations from th prsnt algorithm ar also includd in th tabl. As can b sn from th tabl, running th prsnt algorithm at lowr accuracy improvs th fficincy of th computations and dcrass th computational tim by up to 45%. Compard to othr fficint and low-accuracy algorithms in th litratur [Hui t al. 978; Humlíčk 98; Popp t al. 990; Lthr t al. 99; Shippony t al. 993; Widman 994], th prsnt algorithm sms to b mor rliabl vn at low-accuracy. In addition, othr algorithms fail in som rgions of th computational domain, particularly nar th ral axis (vry small valus of y); for xampl, th Popp and Wijrs algorithm [Popp t al. 990], known for its accuracy, fails in this rgion, rturning rsults for th ral part of th Faddyva function that ar svral ordrs of magnitud away from th corrct valus. Figur shows a comparison btwn th calculations of th partial drivativ V ( x, y) x using th prsnt algorithm (run at th lowst accuracy) and calculations from Algorithm 680, for y=0-0, in th rgion x=[6.-6.5]. Hr w s that th rsults from th prsnt algorithm (vn whn run at th lowst accuracy) sm to b mor accurat and mor rliabl than computations from Popp and Wijrs algorithm in this rgion, whr th lattr loss its accuracy and fails to produc th corrct bhavior of V ( x, y) x. V rf rf 4

15 Figur V(x,y)/ x as calculatd, from th prsnt algorithm (tiny=0-4 ) and from Popp and Wijrs algorithm, using (), for y=0-0. Figur 3 shows a surfac plot of th rlativ rror δ V = ( V Vrf )/ Vrf for th rsults obtaind from Hui s algorithm [Hui t al. 978] using rsults from th prsnt algorithm as rfrnc valus. W not that th Matlab vrsion of Hui s algorithm (crf.m [Hui t al. 978]) mployd in this comparison, uss th p=5 rational approximation whr p is th dgr of nominator polynomial. As is clar from th plot, th rlativ rrors in th rsults of Hui s algorithm ar vry larg for small valus of y and rach 4 ordr of magnitud for mdium valus of x whn y=0-0. In addition to th larg rrors for mdium valus of x and small valus of y, Hui s algorithm producs ngativ valus for th Voigt function (ral part of th Faddyva function) for xampl, for y=0-5 and x=4. Th Voigt function is positiv ovr th whol first quadrant. 5

16 Figur 3 Absolut rlativ rror, δ V = ( V Vrf )/ V, in th calculations of th ral part of th rf Faddyva function from Hui s algorithm using rsults from th prsnt algorithm (tiny=tiny min ) as rfrnc valus. Figur 4 shows a comparison btwn th calculations of th partial drivativ V ( x, y) x using th prsnt algorithm (run at th lowst accuracy) and calculations from Hui s algorithm, using (), for y=0-0, in th rgion x=[7,5]. As w s from th figur, calculations from th prsnt algorithm sm to b mor accurat and mor rliabl than Hui s algorithm which fails to produc th corrct bhavior (ngativ V ( x, y) x ) or th corrct ordr of magnitud of V ( x, y) x in this rgion of th computational domain. W not that th rror contours givn in [Hui t al. 978] wr prsntd ithr for th modulus of th complx rror function or for th absolut valu of th Voigt function V(x,y). That was, probably, th rason why such failurs wr not clar from thir papr. Although Hui s algorithm taks about 0% of th computational tim takn by th prsnt algorithm (for tiny=0-4 ) and about 5% of th computational tim takn by th prsnt algorithm for th highst-accuracy computations, th fact that it fails to produc th corrct valus or corrct signs of th function or vn its corrct bhavior, in this rgion of th computational domain, poss important qustions about its rliability. 6

17 Figur 4 V(x,y)/ x as calculatd, from th prsnt algorithm (tiny=0-4 ) and from Hui s algorithm, using (), for y=0-0. Humlíčk [Humlíčk 98] rportd that any rational approximation suffrs invitabl failur nar th ral axis and h attmptd to ovrcom this failur in his algorithm, w4. Howvr, invstigating th rsults of Humlíčk s algorithm w found that it also suffrs complt failur nar th ral axis in th vicinity of x=5.5. Th rsults for y=0-0 show that Humlíčk s algorithm undrstimats th ral part of th Faddyva function by 8 ordrs of magnitud and by 3 ordrs of magnitud for y=0-5. Figur 5 shows a surfac plot of th rlativ rror in th calculation of V(x,y) using Humlíčk s algorithm taking th rsults from th prsnt algorithm as rfrnc. Tabl 5 shows that th computational tim using th Humlíčk s original cod is almost thr tims that usd by th prsnt algorithm (for th highst-accuracy computations). Evn using a mor fficint vrsion of Humlíčk s cod, modifid by th prsnt authors, w not that th computational tim takn by Humlíčk s algorithm is still longr than that takn by th prsnt algorithm (for th highst-accuracy computations). Figur 6, on th othr hand, shows a comparison btwn th calculations of th partial drivativ V ( x, y) x using () from th prsnt algorithm (run at th lowst accuracy) and thos calculations from Humlíčk s algorithm, for y=0-0, in th rgion x=[5.4,6.4]. This shows that Humlíčk s algorithm dos not produc th corrct bhavior of V ( x, y) x in this domain. 7

18 Figur 5 Absolut rlativ rror, δ V = ( V Vrf )/ Vrf, in th calculations of th ral part of th Faddyva function from Humlíčk s algorithm using rsults from th prsnt algorithm (tiny=tiny min ) as rfrnc valus. Figur 6 V(x,y)/ x as calculatd, from th prsnt algorithm (tiny=0-4 ) and from Humlíčk s algorithm, using (), for y=0-0. 8

19 As with Hui s algorithm, Widman s algorithm [Widman 994] also producs ngativ valus for th ral part of th Faddyva function nar th ral axis. Th ngativ valus for th Voigt function calculatd from Widman s algorithm appar for all valus of th paramtr N (numbr of trms in th rational sris) for y=0-0. A surfac plot of th rlativ rror in th calculations of th ral part of th Faddyva function using Widman s algorithm with N=56 taking th rsults of th prsnt algorithm as a rfrnc is shown in Figur 7. Th figur shows that th rrors rsulting from Widman s algorithm ar catastrophic for small valus of y and that th computd magnitud of V(x,y) is ovrstimatd by up to 6 ordrs of magnitud for y=0-0. Th situation bcoms vn wors for smallr valus of N. Tabl 5 shows that th run tim of th prsnt algorithm at highst accuracy is shortr than th run tim of Widman s algorithm with N=56. Figur 7 Absolut rlativ rror, δ V = ( V Vrf )/ Vrf, in th calculations of th ral part of th Faddyva function from Widman s algorithm, N=56, using rsults from th prsnt algorithm (tiny=tiny min ) as rfrnc valus. Figurs (8-a) and (8-b) show comparisons btwn th calculations of th partial drivativ V ( x, y) x using th prsnt algorithm (run at th lowst accuracy) and Widman s algorithm with N=8 and N=3, rspctivly. Th calculations shown in th figur ar for th rgion x=[6,5] and y=0-0 in Figur (8-a) and y=0-0 in Figur (8- b). From ths figurs, w s that Widman s algorithm dos not rproduc th corrct bhavior for V ( x, y) x in th rgions shown. 9

20 Figur 8-a V(x,y)/ x as calculatd, from th prsnt algorithm (tiny=0-4 ) and from Widman s algorithm with N=8, using (), for y=0-0. Figur 8-b V(x,y)/ x as calculatd, from th prsnt algorithm (tiny=0-4 ) and from Widman s algorithm with N=3, using (), for y=0-0. 0

21 Othr comptitiv algorithms in th litratur also show loss of accuracy in som rgions of th computational domain. For xampl, th algorithm by Shippony and Rad [Shippony t al. 993] xhibits th sam failur in calculating th ral part of th Faddyva function nar th ral axis. In particular w dtctd th sam loss of accuracy suffrd by Popp and Wijrs algorithm for vry small valus of y nar x=6.3. In addition, th Shippony and Rad algorithm producs ngativ valus for V(x,y) and/or L(x,y) in svral rgions of th computational domain. Just for xampl w rfr to th points at x=.5 and y=.5,.0,.5, 3.0, 3.5, 4.0, 5.0 tc. It has to b notd that ths failurs hav bn obtaind vn with th us of th corrction providd by Shippony and Rad in [Shippony t al. 003]. Th situation is no bttr with th Ltchworth and Bnnr s algorithm [Ltchworth t al. 007] whr similar failurs and loss of accuracy ar obtaind for vry small valus of y and valus of x gratr than but clos to x=5.76. For x=5.76 and y=0-0 Ltchwoth and Bnnr s algorithm rturns, for th ral part of Faddyva function th valu V= whil th valu rturnd from th prsnt algorithm is V= and that rturnd from using th function rfi(z) from Mathmatica TM is Th algorithm in [Zaghloul 007] rturns Figur 9 shows a similar comparison btwn th calculations of th partial drivativ V ( x, y) x using th prsnt algorithm (with tiny=0-4 ) and calculations from Ltchworth and Bnnr algorithm, in (), for y=0-0. Th failur of Ltchworth and Bnnr s algorithm for valus of x gratr than but clos to x=5.76 can b asily rcognizd from th figur. It has to b mphasizd hr that ths "comptitiv cods" wr (probably) not dsignd for such xtrm valus of y and th tsts prsntd hr ar mor a dmonstration of th "suprior accuracy" of our softwar vn for xtrm valus. Figur 9 V(x,y)/ x as calculatd, from th prsnt algorithm (tiny=0-4 ) and from Ltchworth & Bnnr s algorithm, using (), for y=0-0.

22 7. CONCLUSIONS An algorithm accompanid by a computr cod, in th form of a MATLAB TM function, for th numrical valuation of th Faddyva function w(z) is prsntd. Th algorithm is mor accurat and avoids failurs discovrd in othr comptitiv publishd algorithms. In addition to its suprior accuracy, th prsnt algorithm and computr cod allow a flxibl accuracy vs fficincy trad-off through controlling a fr paramtr tiny. By adjusting th valu of this paramtr th function can b run with a lowr accuracy and shortr computational tim or high accuracy and longr computational tim. Evn whn run at its lowst accuracy, th prsnt algorithm avoids major problms suffrd by othr comptitiv cods. For all lvls of accuracy th prsnt cod is safr and mor rliabl sinc it dos not rturn ngativ valus for ral and/or imaginary parts of th Faddyva function nor dos it suffr from th loss of accuracy xhibitd by som of th othr comptitiv cods. Th prsnt algorithm can, thrfor, b safly usd and implmntd in prsonal and commrcial libraris.

23 Tabl : Rsults from algorithms in th litratur in comparison with rsults from th prsnt algorithm for som slctd valus of z z Mathmatica TM Algorithm 680 [Zaghloul 007] Prsnt algorithm x y V L V L V V L Fails to valuat rfi $ Fails to valuat rfi Fails to valuat rfi Fails to valuat rfi Fails to valuat rfi * This is th corrct valu as calculatd using (4) in [Zaghloul 007]. Th valu givn in Tabl 4 in th sam rfrnc is calculatd using th asymptotic xprssion for y 0 $ Calculatd using th asymptotic xprssion for y 0 3

24 Tabl 3: Valus of th rlativ rrorsδ V = ( V Vrf ) / Vrf & δ L = ( L Lrf )/ Lrf in calculating th Faddyva function by diffrnt cods using valus of th function calculatd using rfi(z) from Mathmatica as rfrnc valus. z Algorithm 680 Zaghloul [007] Prsnt algorithm x y V L V V L Maximum no. of sris trms

25 Tabl 4: Accuracy vs fficincy trad-off of th prsnt algorithm (computations ar prformd on an array of,840,70 points gnratd using th grid y=logspac(-0, 4, 7) and x=linspac(- 00, 00, 4000) using Matlab (R009b) ). tiny δ V, max = Vtiny Vtiny min / Vtiny δ L, max = Ltiny L L Run tim (s) ε min tiny min / tiny min Tabl 5: Running tims of th prsnt algorithm (for thr valus of th paramtr tiny) compard with othr comptitiv algorithms (computations ar prformd on an array of,840,70 points gnratd using th grid y=logspac(-0, 4, 7) and x=linspac(-00, 00, 4000) using Matlab (R009b)) * Algorithm Run tim (s) Commnts Faddyva, tiny= ε Faddyva, tiny=-8 Faddyva, tiny= Popp & Wijrs [990] Larg rror in th vicinity of x=6.3 & vry small valus of y Humlíčk [98] (original) 3. Larg rror and loss of accuracy in th vicinity Humlíčk [98] (modifid) Widmann [994], N=6 Widmann [994], N=3 Widmann [994], N=64 Widmann [994], N=8 Widmann [994], N= of x=5.6 and vry small valus of y Ngativ valus for V(x,y) nar x-axis Incorrct bhavior and ordr of magnitud of V(x,y)/ x for vry small valus of y. Hui t al [978] 0.4 Larg rror for small valus of y Ngativ valus for V(x,y) (.g. at y=0-5 & x=4) Incorrct bhavior and ordr of magnitud of V(x,y)/ x for vry small valus of y. * Timing rsults dpnd on both hardwar and th vrsion of th softwar usd and can chang significantly. 5

26 ACKNOWLEDGMENTS Th authors would lik to acknowldg valuabl commnts and suggstions rcivd from th rviwrs. In particular th commnts and suggstions rcivd from th associat ditor, algorithm ditor and th fourth anonymous rfr wr xtrmly hlpful and insightful. W would lik also to thank Prof. C. Bnnr from Collg of William and Mary, Williamsburg, VA, USA, for snding us a copy of Ltchworth & Bnnr s computr cod. REFERENCES ABRAMOWITZ M. AND STEGUN, I. A. 97. Handbook of mathmatical functions, Dovr Publications, Inc., Nw York. ABRAROV S.M., QUINE B.M. AND JAGPAL R.K. 00. Rapidly convrgnt sris for highaccuracy calculation of th Voigt function. J. Quant. Spctrosc. & Radiat. Transfr, Vol., ABRAROV S.M., QUINE B.M. AND JAGPAL R.K. 00. High-accuracy approximation of th complx probability functions by Fourir xpansion of xponntial multiplir. Computr Physics Communications, Vol. 8, ARMSTRONG, B.H Spctrum Lin Profils: Th Voigt Function. J. Quant. Spctrosc. & Radiat. Transfr, Vol. 7, 6-88 BOAS M.L Mathmatical Mthods in th Physical Scincs, Third Edition. John Wily & Sons, Inc. NJ, USA. DOMINGUEZ, H.J., LLAMAS, H.F. PRIETO, A.C. AND ORTEGA, A.B A Simpl Rlationship btwn th Voigt Intgral and th Plasma Disprsion Function. Additional Mthods to Estimat th Voigt Intgral. Nuclar Instrumnts and Mthods in Physics Rsarch A. Vol. 78, GAUTSCHI, W Algorithm 363-Complx rror function, Commun. ACM, 635. GAUTSCHI, W Efficint Computation of th Complx Error Function. SIAM J. Numr. Anal., Vol. 7, HUI, A.K., ARMSTRONG, B.H. AND WRAY, A.A Rapid Computation of th Voigt and Complx Error Functions. J. Quant. Spctrosc. & Radiat. Transfr, Vol. 9, HUMLÍČEK J. 98. Optimizd Computation of th Voigt and Complx Probability Functions. J. Quant. Spctrosc. & Radiat. Transfr, Vol. 7, No. 4, LETCHWORTH K.L. AND BENNER D.C 007. Rapid and accurat calculation of th Voigt function. J. Quant. Spctrosc. & Radiat. Transfr, Vol. 07, LETHER F.G. AND WENSTON P.R. 99. Th numrical computation of th Voigt function by a corrctd midpoint quadratur rul for (-, ). Journal of Computational & Applid Mathmatics Vol. 34, 75-9 LUQUE, J. M. CALZADA, M. D. AND SAEZ, M A nw procdur for obtaining th Voigt function dpndnt upon th complx rror function. J. Quant. Spctrosc. & Radiat. Transfr, Vol. 94, 5-6. POPPE, G.P.M. AND C. WIJERS, M. J Mor Efficint Computation of th Complx Error Function. ACM Transactions on Mathmatical Softwar, Vol. 6, No., POPPE G.P.M. AND WIJERS, C.M.J Algorithm 680, Evaluation of th Complx Error Function. ACM Transactions on Mathmatical Softwar, Vol. 6, No., 47. 6

27 SALZER, H.E. 95. Formulas for calculating th rror function of a complx variabl. Math. Tabls and Othr Aids to Computation 5, SCHREIER, F. 99. Th Voigt and Complx Error Function: A Comparison of Computational Mthods. J. Quant. Spctrosc. & Radiat. Transfr, Vol. 48, No. 5/6, SHIPPONY Z. AND READ W.G A highly accurat Voigt function algorithm. J. Quant. Spctrosc. & Radiat. Transfr, Vol. 50, SHIPPONY Z. AND READ W.G A corrction to a highly accurat Voigt function algorithm. J. Quant. Spctrosc. & Radiat. Transfr, Vol. 78,, 55. WEIDEMAN, J.A.C Computation of th Complx Error Function. SIAM J. Numr. Anal. Vol. 3, No. 5, WELLS, R.J Rapid Approximation to th Voigt/Faddva Function and its Drivativs. J. Quant. Spctrosc. & Radiat. Transfr, Vol. 6, WOLFRAM RESEARCH, INC., 008. Mathmatica, Vrsion 7.0, Champaign, IL. ZAGHLOUL, M. R On th calculation of th Voigt Lin-Profil: A singl propr intgral with a dampd sin intgrand. Mon. Not. R. Astron. Soc., Vol. 375, No. 3, ZAGHLOUL, M. R Commnt on: A fast mthod of modling spctral lins. J. Quant. Spctrosc. & Radiat. Transfr, Vol. 09,

AN ANALYSIS OF TELEPHONE MESSAGES: MINIMIZING UNPRODUCTIVE REPLAY TIME

AN ANALYSIS OF TELEPHONE MESSAGES: MINIMIZING UNPRODUCTIVE REPLAY TIME AN ANALYSIS OF TELEPHONE MESSAGES: MINIMIZING UNPRODUCTIVE REPLAY TIME Michal D. Fltwood, Danill L. Paig, Chris S. Fick, and Knnth R. Laughry, Sr. Dpartmnt of Psychology Ric Univrsity Houston, TX flt@ric.du

More information

Reliability Demonstration Test Plan

Reliability Demonstration Test Plan Rliability Dmonstration Tst Plan STATGRAPHICS Cnturion Rv. 6/7/04 Summary... Exampl... Analysis Window... Output... 4 Calculations... 5 Distributions... 5 Summary This procdur crats tst plans to dmonstrat

More information

EXPERIMENTAL DRYING OF TOBACCO LEAVES

EXPERIMENTAL DRYING OF TOBACCO LEAVES 6 TH INTERNATIONAL MULTIDISCIPLINARY CONFERENCE EXPERIMENTAL DRYING OF TOBACCO LEAVES Bndk Krks and Tamás Antal Collg of Nyírgyháza, Faculty of Enginring and Agricultur, H-441 Nyírgyháza, Hungary, E-mail:

More information

PRELIMINARY STUDY ON DISPLACEMENT-BASED DESIGN FOR SEISMIC RETROFIT OF EXISTING BUILDINGS USING TUNED MASS DAMPER

PRELIMINARY STUDY ON DISPLACEMENT-BASED DESIGN FOR SEISMIC RETROFIT OF EXISTING BUILDINGS USING TUNED MASS DAMPER Not: this papr was not abl to b rviwd in accordanc with DEST rquirmnts. PRELIMINARY STUDY ON DISPLACEMENT-BASED DESIGN FOR SEISMIC RETROFIT OF EXISTING BUILDINGS USING TUNED MASS DAMPER Chang-Yu Chn 1

More information

TWO REFERENCE japollo LUNAR PARKING - ORBITS / T. P. TIMER. (NASA CR OR rmx OR AD NUMBER) OCTOBER 1965 GODDARD SPACE FLIGHT CENTER

TWO REFERENCE japollo LUNAR PARKING - ORBITS / T. P. TIMER. (NASA CR OR rmx OR AD NUMBER) OCTOBER 1965 GODDARD SPACE FLIGHT CENTER x-543-55-399 * 1 TWO REFERENCE japollo LUNAR PARKING - ORBITS / I - -. -! BY T. P. TIMER,< CFSTI PRICE(S) $ c 4 (PAGES1 (NASA CR OR rmx OR AD NUMBER) 277 I (CATEGORY) ff 653 July 65 OCTOBER 1965,r ; I

More information

Implementation of a planar coil of wires as a sinusgalvanometer. Analysis of the coil magnetic field

Implementation of a planar coil of wires as a sinusgalvanometer. Analysis of the coil magnetic field mplmntation of a planar coil of wirs as a sinusgalvanomtr Analysis of th coil magntic fild Dimitar G Stoyanov Sofia Tchnical Univrsity, Slivn Enginring and Pdagogical Faculty, 59 Burgasko Shoss Blvd, 88

More information

Going Below the Surface Level of a System This lesson plan is an overview of possible uses of the

Going Below the Surface Level of a System This lesson plan is an overview of possible uses of the Titl Acknowldgmnts Ovrviw Lngth Curriculum Contxt Lsson Objctiv(s) Assssmnt Systms Thinking Concpt(s) Instructional Considrations Matrials Going Blow th Surfac Lvl of a Systm This lsson plan is an ovrviw

More information

PHA Exam 1. Spring 2013

PHA Exam 1. Spring 2013 PHA 5128 Exam 1 Spring 2013 1 Antibiotics (5 points) 2 Body Wight/Pdiatrics (5 points) 3 Rnal Disas (10 points) 4 Aminoglycosids (5 points) 5 Amikacin (10 points) 6 Gntamicin (10 points) 7 Aminoglycosids

More information

Difference in Characteristics of Self-Directed Learning Readiness in Students Participating in Learning Communities

Difference in Characteristics of Self-Directed Learning Readiness in Students Participating in Learning Communities Advancd Scinc and Tchnology Lttrs, pp.135-14 http://dx.doi.org/1.14257/astl.215.92.28 Diffrnc in Charactristics of Slf-Dirctd Larning Radinss in Studnts Participating in Larning Communitis Hur, Young Ju

More information

Blind Estimation of Block Interleaver Parameters using Statistical Characteristics

Blind Estimation of Block Interleaver Parameters using Statistical Characteristics Advancd Scinc and Tchnology Lttrs Vol.139 (FGC 2016), pp.51-56 http://dx.doi.org/10.14257/astl.2016.139.10 Blind Estimation of Block Intrlavr Paramtrs using Statistical Charactristics Jinwoo Jong 1, Youngkyun

More information

Design of a Low Noise Amplifier in 0.18µm SiGe BiCMOS Technology

Design of a Low Noise Amplifier in 0.18µm SiGe BiCMOS Technology Dsign of a Low Nois Amplifir in 0.8µm SiG BiCMOS Tchnology Astract Wi Wang, Fng Hu, Xiaoyuan Bao, Li Chn, Mngjia Huang Chongqing Univrsity of Posts and Tlcommunications, Chongqing 400065, China A 60GHz

More information

MATH 1300: Finite Mathematics EXAM 1 15 February 2017

MATH 1300: Finite Mathematics EXAM 1 15 February 2017 MATH 1300: Finit Mathmatics EXAM 1 15 Fbruary 2017 NAME:... SECTION:... INSTRUCTOR:... SCORE Corrct (A): /15 = % INSTRUCTIONS 1. DO NOT OPEN THIS EXAM UNTIL INSTRUCTED TO BY YOUR ROOM LEADER. All xam pags

More information

Emerging Subsea Networks

Emerging Subsea Networks MODELLING OF NONLINEAR FIBER EFFECTS IN SYSTEMS USING CODIRECTIONAL RAMAN AMPLIFICATION Nlson Costa (Coriant Portugal), Lutz Rapp (Coriant R&D GmbH) Email: nlson.costa@coriant.com Coriant Portugal, R.

More information

FEM Analysis of Welded Spherical Joints Stiffness Fan WANG a, Qin-Kai CHEN b, Qun WANG b, Ke-Wei ZHU b, Xing WANG a

FEM Analysis of Welded Spherical Joints Stiffness Fan WANG a, Qin-Kai CHEN b, Qun WANG b, Ke-Wei ZHU b, Xing WANG a Intrnational Confrnc on Mchanics and Civil Enginring (ICMCE 014) FEM Analysis of Wldd phrical Joints tiffnss Fan WANG a, Qin-Kai CHEN b, Qun WANG b, K-Wi ZHU b, Xing WANG a chool of Architctur and Civil

More information

Design and simulation of the microstrip antenna for 2.4 GHz HM remote control system Deng Qun 1,a,Zhang Weiqiang 2,b,Jiang Jintao 3,c

Design and simulation of the microstrip antenna for 2.4 GHz HM remote control system Deng Qun 1,a,Zhang Weiqiang 2,b,Jiang Jintao 3,c Dsign and simulation of th microstrip antnna for 2.4 GHz HM rmot control systm Dng Qun 1,a,Zhang Wiqiang 2,b,Jiang Jintao 3,c 1,2,3 Institut of Information Enginring &Tchnical, Ningbo Univrsity,Ningbo,

More information

National Assessment in Sweden. A multi-dimensional (ad)venture

National Assessment in Sweden. A multi-dimensional (ad)venture Challngs in Educational Masurmnt Contnt, Mthods and Consquncs Gothnburg, 12 Oct. 2016 National Assssmnt in Swdn A multi-dimnsional (ad)vntur Gudrun Erickson Univrsity of Gothnburg Dpt. of Education and

More information

EXPERIMENT 4 DETERMINATION OF ACCELERATION DUE TO GRAVITY AND NEWTON S SECOND LAW

EXPERIMENT 4 DETERMINATION OF ACCELERATION DUE TO GRAVITY AND NEWTON S SECOND LAW EXPERIMENT 4 DETERMINATION OF ACCELERATION DUE TO GRAVITY AND NEWTON S SECOND LAW I. Introduction. Thr ar two objctivs in this laboratory xrcis. Th first objctiv, (A), is th study of th bhavior of a body

More information

Form. Tick the boxes below to indicate your change(s) of circumstance and complete the relevant sections of this form

Form. Tick the boxes below to indicate your change(s) of circumstance and complete the relevant sections of this form tification of chang of circumstancs for EU studnts on full-tim courss - Acadmic Yar 2013/14 Form EUCO1 This form is also availabl at www.gov.uk/studntfinanc First nam(s) Surnam/family nam Important information

More information

Reliability of fovea palatinea in determining the posterior palatal seal

Reliability of fovea palatinea in determining the posterior palatal seal J Bagh Collg Dntistry Vol.21(1, 9 Rliability of fova Rliability of fova palatina in dtrmining th postrior palatal sal Yasmn T. AL Alousi, B.D.S, M.Sc. (1 ABSTRACT Background: Th prsnt study was carrid

More information

Fall 2005 Economics and Econonic Methods Prelim. (Shevchenko, Chair; Biddle, Choi, Iglesias, Martin) Econometrics: Part 4

Fall 2005 Economics and Econonic Methods Prelim. (Shevchenko, Chair; Biddle, Choi, Iglesias, Martin) Econometrics: Part 4 Fall 2005 Economics and Econonic Mthods Prlim (Shvchnko, Chair; Biddl, Choi, Iglsias, Martin) Economtrics: Part 4 Dirctions: Answr all qustions. Point totals for ach qustion ar givn in parnthsis; thr ar

More information

Cattle Finishing Net Returns in 2017 A Bit Different from a Year Ago Michael Langemeier, Associate Director, Center for Commercial Agriculture

Cattle Finishing Net Returns in 2017 A Bit Different from a Year Ago Michael Langemeier, Associate Director, Center for Commercial Agriculture May 2017 Cattl Finishing Nt Rturns in 2017 A Bit Diffrnt from a Yar Ago Michal Langmir, Associat Dirctor, Cntr for Commrcial Agricultur With th xcption of May 2016, monthly fd cattl nt rturns wr ngativ

More information

Hybrid force-position control for manipulators with 4 degrees of freedom

Hybrid force-position control for manipulators with 4 degrees of freedom Hybrid forc-position control for manipulators with 4 dgrs of frdom Alxandru GAL Institut of Solid Mchanics of th Romanian Acadmy C-tin Mill 5, Bucharst, Romania galxandru@yahoo.com Abstract: his papr taks

More information

Optimize Neural Network Controller Design Using Genetic Algorithm

Optimize Neural Network Controller Design Using Genetic Algorithm Procdings of th 7th World Congrss on Intllignt Control and Automation Jun 25-27, 28, Chongqing, China Optimiz Nural Ntwork Controllr Dsign Using Gntic Algorithm Aril Kopl, Xiao-Hua Yu Dpartmnt of Elctrical

More information

AGE DETERMINATION FROM RADIOLOGICAL STUDY OF EPIPHYSIAL APPEARANCE AND FUSION AROUND ELBOW JOINT *Dr. S.S. Bhise, **Dr. S. D.

AGE DETERMINATION FROM RADIOLOGICAL STUDY OF EPIPHYSIAL APPEARANCE AND FUSION AROUND ELBOW JOINT *Dr. S.S. Bhise, **Dr. S. D. AGE DETERMINATION FROM RADIOLOGICAL STUDY OF EPIPHYSIAL APPEARANCE AND FUSION AROUND ELBOW JOINT *Dr. S.S. Bhis, **Dr. S. D. Nanandkar * Corrsponding author, Assistant profssor, Fornsic mdicin dpt., Grant

More information

Measuring Cache and TLB Performance and Their Effect on Benchmark Run Times

Measuring Cache and TLB Performance and Their Effect on Benchmark Run Times Masuring Cach and TLB Prformanc and Thir Effct on Bnchmark Run Tims Rafal H. Saavdra Alan Jay Smith ABSTRACT In prvious rsarch, w hav dvlopd and prsntd a modl for masuring machins and analyzing programs,

More information

Research into the effect of the treatment of the carpal tunnel syndrome with the Phystrac traction device

Research into the effect of the treatment of the carpal tunnel syndrome with the Phystrac traction device Rsarch into th ffct of th tratmnt of th carpal tunnl syndrom with th Phystrac traction dvic Rsarch carrid out in commission of: Fysiothrapi Cntrum Zuidwold By: Irn Kloostrman MA Octobr 2006 Forword This

More information

THEORY OF ACOUSTIC EMISSION FOR MICRO-CRACKS APPEARED UNDER THE SURFACE LAYER MACHINING BY COMPRESSED ABRASIVE

THEORY OF ACOUSTIC EMISSION FOR MICRO-CRACKS APPEARED UNDER THE SURFACE LAYER MACHINING BY COMPRESSED ABRASIVE THEORY OF ACOUSTIC EMISSION FOR MICRO-CRACKS APPEARED UNDER THE SURFACE LAYER MACHINING BY COMPRESSED ABRASIVE A.K. Aringazin, 1, V.D. Krvchik,, V.A. Skryabin, M.B. Smnov,, G.V. Tarabrin 1 Eurasian National

More information

A Practical System for Measuring Film Thickness. Means of Laser Interference with Laminar-Like Laser

A Practical System for Measuring Film Thickness. Means of Laser Interference with Laminar-Like Laser A Practical Systm for Masuring Film Thicknss by Mans of Lasr Intrfrnc with Laminar-Lik Lasr Fng ZHU, Kazuhiko ISHIKAWA, Toru IBE, Katsuhiko ASADA,' and Masahiro UEDA4 Dpartmnt of Information Scinc, Faculty

More information

e/m apparatus (two similar, but non-identical ones, from different manufacturers; we call them A and B ) meter stick black cloth

e/m apparatus (two similar, but non-identical ones, from different manufacturers; we call them A and B ) meter stick black cloth Stony Brook Physics Laboratory Manuals Lab 6 - / of th lctron Th purpos of this laboratory is th asurnt of th charg ovr ass ratio / for th lctron and to study qualitativly th otion of chargd particls in

More information

Components Required: Small bread-board to build the circuit on( or just use clip leads directly) 2ea 220pF capacitors 1 ea 1nF 10uH inductor

Components Required: Small bread-board to build the circuit on( or just use clip leads directly) 2ea 220pF capacitors 1 ea 1nF 10uH inductor EELE445 Lab 3: Whit nois, ½H(f)½, and a x3 Frquncy Multiplir Purpos Th purpos of th lab is to bcom acquaintd with PSD, whit nois and filtrs in th tim domain and th frquncy domain. Whit nois and swpt sin

More information

REGRESSION ASSOCIATION VS. PREDICTION

REGRESSION ASSOCIATION VS. PREDICTION BIOSTATISTICS WORKSHOP: REGRESSION ASSOCIATION VS. PREDICTION Sub-Saharan Africa CFAR mting July 18, 2016 Durban, South Africa Rgrssion what is it good for? Explor Associations Btwn outcoms and xposurs

More information

MUDRA PHYSICAL SCIENCES

MUDRA PHYSICAL SCIENCES Physical Scincs For ET & SET Exams. Of UGC-CSIR MUDRA PHYSICAL SCIECES VOLUME-05 PART B & C MODEL QUESTIO BAK FOR THE TOPICS: 7. Exprimntal Tchniqus and Data Analysis UIT-I UIT-II 5 UIT-III 9 8. Atomic

More information

Statistical Techniques For Comparing ACT-R Models of Cognitive Performance

Statistical Techniques For Comparing ACT-R Models of Cognitive Performance Statistical Tchniqus For Comparing ACT-R Modls of Cognitiv Prformanc Ryan Shaun Bakr (rsbakr@cmu.du) Albrt T. Corbtt (corbtt@cmu.du) Knnth R. Kodingr (kodingr@cmu.du) Human-Computr Intraction Institut,

More information

Time Variation of Expected Returns on REITs: Implications for Market. Integration and the Financial Crisis

Time Variation of Expected Returns on REITs: Implications for Market. Integration and the Financial Crisis Tim Variation of Expctd Rturns on REITs: Implications for Markt Intgration and th Financial Crisis Author Yuming Li Abstract This articl uss a conditional covarianc-basd thr-factor pricing modl and a REIT

More information

Labyrinth Seal Design Optimization Based on Quadratic Regression Orthogonal Experiment

Labyrinth Seal Design Optimization Based on Quadratic Regression Orthogonal Experiment Enrgy and Powr Enginring, 2017, 9, 204-215 http://www.scirp.org/ournal/p ISSN Onlin: 1947-3818 ISSN Print: 1949-243X Labyrinth Sal Dsign Optimization Basd on Quadratic Rgrssion Orthogonal Exprimnt Lihua

More information

Evaluation Of Logistic Regression In Classification Of Drug Data In Kwara State

Evaluation Of Logistic Regression In Classification Of Drug Data In Kwara State Intrnational Journal Of Computational Enginring Rsarch (icronlin.com) Vol. 3 Issu. 3 Evaluation Of Logistic Rgrssion In Classification Of Drug Data In Kwara Stat, O.S. Balogun, 2 T.J. Aingbad, A.A. Ainrfon

More information

Evaluation of Accuracy of U.S. DOT Rail-Highway Grade Crossing Accident Prediction Models

Evaluation of Accuracy of U.S. DOT Rail-Highway Grade Crossing Accident Prediction Models 166 TRANSPORTATION RESEARCH RECORD 1495 Evaluation of Accuracy of U.S. DOT Rail-Highway Grad Crossing Accidnt Prdiction Modls M.I. MUTABAZI AND W.D. BERG Svral vrsions of th U.S. Dpartmnt of Transportation

More information

Localization Performance of Real and Virtual Sound Sources

Localization Performance of Real and Virtual Sound Sources M.Sc.E.E. Jan Abildgaard Pdrsn AM3D A/S Riihimäkivj 6 DK-9200 Aalborg Dnmark M.Sc.E.E. Torbn Jørgnsn Trma A/S Hovmarkn 4 DK-8520 Lystrup Dnmark E-mail: jap@am3d.com / toj@trma.dk ABSTRACT This papr dscribs

More information

Rudolf Huber GmbH ELECTROMAGNETIC TOOTH CLUTCHES

Rudolf Huber GmbH ELECTROMAGNETIC TOOTH CLUTCHES Rudolf Hubr GmbH ELECTROMAGNETIC TOOTH CLUTCHES Aubingrwg 41 82178 Puchhim Tl: +49 (0)89 89026426 Fax: +49 (0)89 89026427 www.mz-kupplungn.d info@hubr-prazisionsmchanik.d Elctromagntic tooth clutchs with

More information

Combined use of calcipotriol solution (SOp.g/ ml) and Polytar liquid in scalp psoriasis.

Combined use of calcipotriol solution (SOp.g/ ml) and Polytar liquid in scalp psoriasis. MCS9506INT 27 April1999 Pag 11 of 189 Summary This documnt has bn dov.;nloadd from \v\vw.lo-pharma.com subjct to th trms of us stat on th wbsit. It contains data and rsults rgarding approvd and non-approvd

More information

Probability, Genetics, and Games

Probability, Genetics, and Games " Probability, Gntics, and Gams Hav you vr hard of gns? (W don t man th kind you war!) What color ar your ys? Can you curl your tongu? Your birth parnts gav you a uniqu st of gns that dtrmin such things.

More information

EVALUATION OF DIAGNOSTIC PERFORMANCE USING PARTIAL AREA UNDER THE ROC CURVE. Hua Ma. B.S. Sichuan Normal University, Chengdu, China, 2007

EVALUATION OF DIAGNOSTIC PERFORMANCE USING PARTIAL AREA UNDER THE ROC CURVE. Hua Ma. B.S. Sichuan Normal University, Chengdu, China, 2007 EVALUATION OF DIAGNOSTIC PERFORMANCE USING PARTIAL AREA UNDER THE ROC CURVE by Hua Ma B.S. Sichuan Normal Univrsity, Chngdu, China, 2007 M.S. Xiamn Univrsity, Xiamn, China, 2010 Submittd to th Graduat

More information

THE CROSS-FLOW DRAG ON A MANOEUVRING SHIP. J. P. HOOFf. MARIN, Wageningen, The Netherlands

THE CROSS-FLOW DRAG ON A MANOEUVRING SHIP. J. P. HOOFf. MARIN, Wageningen, The Netherlands f!j Prgamon Ocan Engng, Vol. 21, No. 3, pp. 329-342, 1994 Elsvir Scinc Ltd Printd in Grat Britain 0029-8018194$7.00 +.00 THE CROSS-FLOW DRAG ON A MANOEUVRNG SHP J. P. HOOFf MARN, Wagningn, Th Nthrlands

More information

Brushless DC motor speed control strategy of simulation research

Brushless DC motor speed control strategy of simulation research Brushlss DC motor spd control stratgy of simulation rsarch Xiang WEN 1,*,Zhn-qiang LI 2 1,2 Collg of Elctrical and Information Enginring, Guangxi Univrsity of Scinc and Tchnology, Liuzhou Guangxi 55006,

More information

IBM Research Report. A Method of Calculating the Cost of Reducing the Risk Exposure of Non-compliant Process Instances

IBM Research Report. A Method of Calculating the Cost of Reducing the Risk Exposure of Non-compliant Process Instances RC24930 (W1001-025) January 8, 2010 Computr Scinc IBM Rsarch Rport A Mthod of Calculating th Cost of Rducing th Risk Exposur of Non-compliant Procss Instancs Yurdar N. Doganata, Francisco Curbra IBM Rsarch

More information

How to Combine Expert (or Novice) Advice when Actions Impact the Environment?

How to Combine Expert (or Novice) Advice when Actions Impact the Environment? How to Combin Exprt (or Novic) Advic whn Actions Impact th Environmnt? Danila Pucci d Farias Dpartmnt of Mchanical Enginring Massachustts Institut of Tchnology Cambridg, MA 02139 pucci@mit.du Nimrod Mgiddo

More information

Alternate Mount and Location for a Trolling Motor. Print in Landscape Mode with ¼ inch borders.

Alternate Mount and Location for a Trolling Motor. Print in Landscape Mode with ¼ inch borders. SIDE MOTOR MOUNT Drawn 09-15-2013 Altrnat Mount and Location for a Trolling Motor Rv. 09-21-2013 Print in Landscap Mod with ¼ inch bordrs. Th primary purpos of locating th trolling motor nxt to th oprator

More information

A Comment on Variance Decomposition and Nesting Effects in Two- and Three-Level Designs

A Comment on Variance Decomposition and Nesting Effects in Two- and Three-Level Designs DISCUSSION PAPER SERIES IZA DP No. 3178 A Commnt on Varianc Dcomposition and Nsting Effcts in Two- and Thr-Lvl Dsigns Spyros Konstantopoulos Novmbr 007 Forschungsinstitut zur Zukunft dr Arbit Institut

More information

Alternate Mount and Location for a Trolling Motor. Print in Landscape Mode with ¼ inch borders.

Alternate Mount and Location for a Trolling Motor. Print in Landscape Mode with ¼ inch borders. SIDE MOTOR MOUNT Altrnat Mount and Location for a Trolling Motor Drawn 09-15-2013 Rv. 07-11-2016 Print in Landscap Mod with ¼ inch bordrs. Th primary purpos of locating th trolling motor nxt to th oprator

More information

Approximate Dimension Equalization in Vector-based Information Retrieval

Approximate Dimension Equalization in Vector-based Information Retrieval Approximat Dimnsion qualization in Vctor-basd Information Rtrival Fan Jiang Dpartmnt of Computr Scinc, Duk Univrsity, Durham, NC 27708 USA ichal L. Littman AT&T Labs Rsarch, Florham Park, NJ 07932-0971

More information

APPLYING THE MIXED RASCH MODEL TO THE FRACTION CONCEPT OF PUPILS

APPLYING THE MIXED RASCH MODEL TO THE FRACTION CONCEPT OF PUPILS Intrnational Journal of Innovativ Managmnt, Information & Production ISME Intrnationalc200 ISSN 285-5439 Volum, Numbr, Dcmbr 200 PP. 90-96 APPLYING THE MIXED RASCH MODEL TO THE FRACTION CONCEPT OF PUPILS

More information

Tests on a Single Phase Transformer

Tests on a Single Phase Transformer Hong Kong nstitut of ational Education (Tsing Yi) Dpartmnt of Enginring Elctrical Enginring Principls Laboratory Sht: EEE3405/LAB03 Studnt nam: Cours / Yar: Dat: Tsts on a Singl Phas Transformr Objctivs

More information

A multiple mediator model: Power analysis based on Monte Carlo simulation

A multiple mediator model: Power analysis based on Monte Carlo simulation Amrican Journal of Applid Psychology 2014; 3(3): 72-79 Publishd onlin Jun 20, 2014 (http://wwwscincpublishinggroupcom/j/ajap) doi: 1011648/jajap2014030315 A multipl mdiator modl: Powr analysis basd on

More information

Eugene Charniak and Eugene Santos Jr. Department of Computer Science Brown University Providence RI and

Eugene Charniak and Eugene Santos Jr. Department of Computer Science Brown University Providence RI and From: AAAI-92 Procdings. Copyright 1992, AAAI (www.aaai.org). All rights rsrvd. mic MAP Calcul Eugn Charniak and Eugn Santos Jr. Dpartmnt of Computr Scinc Brown Univrsity Providnc RI 02912 c@cs.brown.du

More information

Mathematical Simulation on Self-tuning Fuzzy Controller for Small Cylindrical Object Navigating near Free-surface

Mathematical Simulation on Self-tuning Fuzzy Controller for Small Cylindrical Object Navigating near Free-surface Availabl onlin at www.scincdirct.com Procdia Enginring () 9 96 SREE Confrnc on Enginring Modlling and Simulation (CEMS ) Mathmatical Simulation on Slf-tuning Fuzzy Controllr for Small Cylindrical Objct

More information

Efficient Spectral Power Estimation on an Arbitrary Frequency Scale

Efficient Spectral Power Estimation on an Arbitrary Frequency Scale 178 F. ZPLT, M. KSL, EFFCET SPECTL POWE ESTMTO O BTY FEQUECY SCLE Efficint Spctral Powr Estimation on an rbitrary Frquncy Scal Filip ZPLT, Miroslav KSL Dpt. of adio Elctronics, Brno Univrsity of Tchnology,

More information

A e C l /C d. S j X e. Z i

A e C l /C d. S j X e. Z i DESIGN MODIFICATIONS TO ACHIEVE LOW-BOOM AND LOW-DRAG SUPERSONIC CONCEPTUAL DESIGNS Danil. B. L Advisor: Prof. Jams C. McDanil Univrsity of Virginia, Charlottsvill, VA 94 NASA Mntor: Dr. Wu Li NASA Langly

More information

Efficient MBS-FEM integration for structural dynamics

Efficient MBS-FEM integration for structural dynamics Th 2012 World Congrss on Advancs in Civil, Environmntal, and Matrials Rsarch (ACEM 12) Soul, Kora, August 26-30, 2012 Efficint MBS-FEM intgration for structural dynamics *Dragan Z. Marinkovic 1) and Manfrd

More information

INVESTIGATION OF BOUNDARY LAYER FOR A SECOND ORDER EQUATION UNDER LOCAL AND NON LOCAL BOUNDARY CONDITIONS

INVESTIGATION OF BOUNDARY LAYER FOR A SECOND ORDER EQUATION UNDER LOCAL AND NON LOCAL BOUNDARY CONDITIONS ي J Basic Appl Sci Rs 375-757 TtRoad Publication ISSN 9-434 Journal of Basic and Applid Scintific Rsarch wwwttroadcom INVESTIGATION OF BOUNDARY LAYER FOR A SEOND ORDER EQUATION UNDER LOAL AND NON LOAL

More information

CALCULATION OF INDUCTION DEVICE WITH SIMULATION METHODS

CALCULATION OF INDUCTION DEVICE WITH SIMULATION METHODS ., 53,.I,,, 2010 ANNUAL of th Univrsity of Mining and Gology St. Ivan Rilski, Vol. 53, Part, Mchanization, lctrification and automation in mins, 2010 CALCULATION OF INDUCTION DEVICE WITH SIMULATION METHODS

More information

How Asset Maintenance Strategy Selection Affects Defect Elimination, Failure Prevention and Equipment Reliability

How Asset Maintenance Strategy Selection Affects Defect Elimination, Failure Prevention and Equipment Reliability Availability P +61 (0) 402 731 563 F +61 (8) 9457 8642 E info@liftim-rliability.com How Asst aintnanc Stratgy Slction Affcts Dfct Elimination, Failur Prvntion and Equipmnt Rliability ABSTRACT: Th 20 th

More information

Car Taxes and CO 2 emissions in EU. Summary. Introduction. Author: Jørgen Jordal-Jørgensen, COWI

Car Taxes and CO 2 emissions in EU. Summary. Introduction. Author: Jørgen Jordal-Jørgensen, COWI Car Taxs and CO 2 missions in EU Author: Jørgn Jordal-Jørgnsn, COWI Summary Th ful fficincy of passngr cars is oftn mphasisd as on of th most significant aras of action in trms of limiting th transport

More information

LINE ENHANCER METHODS FOR CARRIER TRACKING IN QAM/PSK DATA SIGNALS

LINE ENHANCER METHODS FOR CARRIER TRACKING IN QAM/PSK DATA SIGNALS LIE EHACER METHODS FOR CARRIER TRACKIG I QAM/PSK DATA SIGALS Randall Flint (Univrsity of Utah, Salt Lak City, Utah, USA; rkflint@sisna.com); Bhrou Farhang-Boroujny (Univrsity of Utah, Salt Lak City, Utah,

More information

YOUR VIEWS ABOUT YOUR HIGH BLOOD PRESSURE

YOUR VIEWS ABOUT YOUR HIGH BLOOD PRESSURE YOUR VIEWS ABOUT YOUR HIGH BLOOD PRESSURE W ar intrstd in your viws about your high blood prssur. Ths ar statmnts othr popl hav mad about thir high blood prssur. Plas show how much you or dis with ach

More information

A Robust R-peak Detection Algorithm using Wavelet Packets

A Robust R-peak Detection Algorithm using Wavelet Packets Intrnational Journal of Computr Applications (975 8887) A Robust R-pak Dtction Algorithm using Wavlt Packts Omkar Singh School of Elctronics and Communication Enginring Lovly Profssional Univrsity Punjab-INDIA

More information

Chemometrics. Derivatives in Spectroscopy. Part I The Behavior of the Derivative. Howard Mark and Jerome Workman Jr.

Chemometrics. Derivatives in Spectroscopy. Part I The Behavior of the Derivative. Howard Mark and Jerome Workman Jr. Chmomtrics Drivativs in Spctroscopy Part I Th Bhavior of th Drivativ Howar Mark an Jrom Workman Jr. Jrom Workman Jr. srvs on th Eitorial Avisory Boar of Spctroscopy an is vic-prsint of rsarch for Argos

More information

Comparison of lower-hybrid (LH) frequency spectra between at the high-field side (HFS) and low-field side (LFS) in Alcator C-Mod

Comparison of lower-hybrid (LH) frequency spectra between at the high-field side (HFS) and low-field side (LFS) in Alcator C-Mod Comparison of lowr-hybrid (LH) frquncy spctra btwn at th high-fild sid (HFS) and low-fild sid (LFS) in Alcator C-Mod S. G. Bak, R. R. Parkr, S. Shiraiwa, G. M. Wallac, P. T. Bonoli, D. Brunnr, I. Faust,

More information

New Methods for Modeling Reliability Using Degradation Data

New Methods for Modeling Reliability Using Degradation Data Papr 263-26 Nw Mthods for Modling Rliability Using Dgradation Data José G. Ramírz, W.L. Gor & Associats, Inc., Elkton, MD Gordon Johnston, SAS Institut, Inc., Cary, NC ABSTRACT Enginrs and rsarchrs us

More information

Audio Engineering Society Convention Paper Presented at the 111th Convention 2001 September New York, NY, USA

Audio Engineering Society Convention Paper Presented at the 111th Convention 2001 September New York, NY, USA Audio Enginring Socity Convntion Papr Prsntd at th th Convntion 200 Sptmbr 2 24 Nw York, NY, USA This convntion papr has bn rproducd from th author's advanc manuscript, without diting, corrctions, or considration

More information

Effective Subgrade Coefficients for Seismic Performance Assessment of Pile Foundations

Effective Subgrade Coefficients for Seismic Performance Assessment of Pile Foundations Effctiv Subgrad Cofficints for Sismic Prformanc Assssmnt of Pil Foundations W.L. Tan, S.T. Song & W.S. Hung National Chung-Hsing Unuvrsity, Taiwan,.O.C. SUMMAY: ( Th soil subgrad cofficints availabl in

More information

Statistical Magnitude Analysis and Distance Determination of the Nearby F8V Stars

Statistical Magnitude Analysis and Distance Determination of the Nearby F8V Stars Enginring, Tchnology & Applid Scinc Rsarch ol. 4, No. 4, 4, 68-685 68 Statistical Magnitud Analysis and Distanc Dtrmination of th Narby F8 Stars Hany R. Dwidar Astronomy, Mtorology and Spac Scinc Dpt.

More information

Hybrid Control Strategy of Vehicle Semi-active Suspension Based on the Genetic Algorithm Optimization

Hybrid Control Strategy of Vehicle Semi-active Suspension Based on the Genetic Algorithm Optimization Hybrid Control Stratgy of Vhicl Smi-activ Suspnsion Basd on th Gntic Algorithm Optimization Quanmin Guo 1, 2 1 School of Mchanical and Prcision Instrumnt Enginring, Xi an Univrsity of chnology, Xi an 710048,

More information

Company registration number: ROI FRS 105 Demo Client UNAUDITED FINANCIAL STATEMENTS for the year ended 31 January 2018

Company registration number: ROI FRS 105 Demo Client UNAUDITED FINANCIAL STATEMENTS for the year ended 31 January 2018 Company rgistration numbr: 999955 UNAUDITED FINANCIAL STATEMENTS for th yar ndd 31 January 2018 Unauditd Financial Statmnts CONTENTS PAGE Dirctors and Othr Information 1 Dirctor s Rport 2 Accountant s

More information

Machine Learning Approach to Identifying the Dataset Threshold for the Performance Estimators in Supervised Learning

Machine Learning Approach to Identifying the Dataset Threshold for the Performance Estimators in Supervised Learning Machin Larning Approach to Idntifying th Datast Thrshold for th Prformanc Estimators in Suprvisd Larning Zanifa Omary, Frdrick Mtnzi Dublin Institut of Tchnology, Irland zanifa.omary@studnt.dit.i, frdrick.mtnzi@dit.i

More information

GIVING FLEXIBILITY TO THE NELSON-SIEGEL CLASS OF TERM STRUCTURE MODELS *

GIVING FLEXIBILITY TO THE NELSON-SIEGEL CLASS OF TERM STRUCTURE MODELS * RESUMO GIVING FLEXIBILITY TO THE NELSON-SIEGEL CLA OF TERM STRUCTURE MODELS * Rafal B. Rznd Est artigo compara as habilidads d intrpolação da strutura a trmo d modlos nãoparamtricos paramétricos, utilizados

More information

Catriona Crossan Health Economics Research Group (HERG), Brunel University

Catriona Crossan Health Economics Research Group (HERG), Brunel University MAPGuid: Modlling of clinical pathways to assss cost-ffctivnss in NICE guidlins: impact on stakholdr viws of th importanc of potntial updat topics Catriona Crossan Halth Economics Rsarch Group (HERG),

More information

AN INVESTIGATION INTO THE EXTENT OF NON-COMPLIANCE WITH THE NATIONAL MINIMUM WAGE 1

AN INVESTIGATION INTO THE EXTENT OF NON-COMPLIANCE WITH THE NATIONAL MINIMUM WAGE 1 AN INVESTIGATION INTO THE EXTENT OF NON-COMPLIANCE WITH THE NATIONAL 1 Stphanus l Roux a, Paolo Lucchino b and David Wilkinson b a) Dpartmnt for Work and Pnsions 2 b) National Institut of Economic and

More information

SOLUTIONS FOR THEORETICAL COMPETITION

SOLUTIONS FOR THEORETICAL COMPETITION VI Intrnational Zhautykov Olympiad Thortical Comptition/Solutions Pag /5 SOLUTIONS FOR THEORETICAL COMPETITION Thortical Qustion A Potntial nrgy of th rigid rod U=mgl/sinα transforms to th kintic nrgy

More information

An explicit unconditionally stable numerical solution of the advection problem in irrotational flow fields

An explicit unconditionally stable numerical solution of the advection problem in irrotational flow fields WATER RESOURCES RESEARCH, VOL. 40, W06501, doi:10.109/003wr00646, 004 An xplicit unconditionally stabl numrical solution of th advction problm in irrotational flow filds Alssandra Bascià Dpartmnt of Mchanics

More information

Policy Coherence for Development In Fisheries and Aquaculture

Policy Coherence for Development In Fisheries and Aquaculture Policy Cohrnc for Dvlopm In Fishris and Aquacultur 10-11 April 2014 Dr Christoph Béné Institut of Dvlopm Studis, UK commissiond by UK-DFID Clustr Quality Evaluation Siz Consistnc y Gap Fish and nutritional

More information

FITTING ELECTRICITY MARKET MODELS. A CONJECTURAL VARIATIONS APPROACH

FITTING ELECTRICITY MARKET MODELS. A CONJECTURAL VARIATIONS APPROACH FITTING LCTRICITY MARKT MODLS. A CONJCTURAL VARIATIONS APPROACH Antonio García-Alcald Mariano Vntosa Michl Rivir Andrés Ramos Grgorio Rlaño INSTITUTO D INVSTIGACIÓN TCNOLÓGICA Univrsidad Pontificia Comillas

More information

Code_Aster. Finite element method isoparametric

Code_Aster. Finite element method isoparametric Cod_Astr Vrsion dfault Titr : La méthod ds élémnts finis isoparamétriqus Dat : 09/10/2013 Pag : 1/20 Rsponsabl : ABBAS Mickaël Clé : R3.01.00 Révision : Finit lmnt mthod isoparamtric Summary: This documnt

More information

UNCERTAINTY IN THE TAYLOR RULE AND MONETARY POLICY ASSESSMENT*

UNCERTAINTY IN THE TAYLOR RULE AND MONETARY POLICY ASSESSMENT* UNCERTAINTY IN THE TAYLOR RULE AND MONETARY POLICY ASSESSMENT* Frnando Martins** Paulo Soars Estvs** 1. INTRODUCTION In rcnt yars, on has witnssd a widsprad attntion on th way montary policy is conductd

More information

Magnetic Field Exposure Assessment of Lineman Brain Model during Live Line Maintenance

Magnetic Field Exposure Assessment of Lineman Brain Model during Live Line Maintenance Procdings of th 14 th Intrnational Middl ast Powr Systms Confrnc (MPCON 10), Cairo Univrsity, gypt, Dcmbr 19-1, 010, Papr ID 109 Magntic Fild xposur Assssmnt of Linman rain Modl during Liv Lin Maintnanc

More information

Code_Aster. Finite element method isoparametric

Code_Aster. Finite element method isoparametric Titr : La méthod ds élémnts finis isoparamétriqus Dat : 10/01/2011 Pag : 1/18 Finit lmnt mthod isoparamtric Abstract : This documnt prsnts th bass of th finit lmnts isoparamtric introducd into for th modlization

More information

A LOW COST COMPUTATION FOR APPROXIMATE PREDICITON OF GAS CORE CROSS SECTIONS IN GAS ASSISTED INJECTION MOLDING.

A LOW COST COMPUTATION FOR APPROXIMATE PREDICITON OF GAS CORE CROSS SECTIONS IN GAS ASSISTED INJECTION MOLDING. A LOW COST COMPUTATION FOR APPROXIMATE PREDICITON OF GAS CORE CROSS SECTIONS IN GAS ASSISTED INJECTION MOLDING. A. Polynkin, J. F. T. Pittman, and J. Sinz Cntr for Polymr Procssing Simulation and Dsign,

More information

2 Arrange the following angles in order from smallest to largest. A B C D E F. 3 List the pairs of angles which look to be the same size.

2 Arrange the following angles in order from smallest to largest. A B C D E F. 3 List the pairs of angles which look to be the same size. I n rcnt yars thr has bn an xplosion in rsarch basd on dinosaur tracks. Using trackways w can tll whthr a dinosaur was walking, trotting, running or wading. W can stimat its spd by looking at th lngth

More information

A NEW TESTING METHOD FOR CREEP BEHAVIOR OF SELF-COMPACTING CONCRETE AT EARLY AGE

A NEW TESTING METHOD FOR CREEP BEHAVIOR OF SELF-COMPACTING CONCRETE AT EARLY AGE A NEW TESTING METHOD FOR CREEP BEHAVIOR OF SELF-COMPACTING CONCRETE AT EARLY AGE Krittiya Kawman A dissrtation submittd to Kochi Univrsity of Tchnology in partial fulfillmnt of th rquirmnts for Th Dgr

More information

Chapter 12 Student Lecture Notes 12-1

Chapter 12 Student Lecture Notes 12-1 Chaptr 1 Studnt Lctur Nots 1-1 Businss Statistics: A Dcision-Making Approach 6 th Edition Chaptr 1 Goodnss-of-Fit Tsts and Contingncy Analysis 005 Prntic-Hall, Inc. Chap 1-1 Chaptr Goals Aftr complting

More information

Sensitivity Analysis of the JPALS Shipboard Relative GPS Measurement Quality Monitor

Sensitivity Analysis of the JPALS Shipboard Relative GPS Measurement Quality Monitor Snsitivity Analysis of th JPALS Shipboard Rlativ GPS Masurmnt Quality Monitor Michal Konig, Dmoz Gbr-Egziabhr, Sam Pulln, Ung-Souk Kim, and Pr Eng Stanford Univrsity Boris S. Prvan and Fang Chng Chan Dpartmnt

More information

Using the Aggregate Demand-Aggregate Supply Model to Identify Structural. Demand-Side and Supply-Side Shocks: Results Using a Bivariate VAR

Using the Aggregate Demand-Aggregate Supply Model to Identify Structural. Demand-Side and Supply-Side Shocks: Results Using a Bivariate VAR Octobr 4, 3 Using th Aggrgat Dmand-Aggrgat Supply Modl to Idntify Structural Dmand-Sid and Supply-Sid Shocks: Rsults Using a Bivariat VAR Jams Pry Covr Univrsity of Alabama Waltr Endrs Univrsity of Alabama

More information

Company registration number: ROI FRS 105 Demo Client UNAUDITED FINANCIAL STATEMENTS for the year ended 31 December 2017

Company registration number: ROI FRS 105 Demo Client UNAUDITED FINANCIAL STATEMENTS for the year ended 31 December 2017 Company rgistration numbr: 999955 UNAUDITED FINANCIAL STATEMENTS for th yar ndd 31 Dcmbr 2017 Profit and Loss Account Not Turnovr 422,560 383,200 Othr incom 1,000-423,560 383,200 Cost of raw matrials and

More information

Simulation of Communication Systems

Simulation of Communication Systems Simulation of Communication Systms By Xiaoyuan Wu Thsis submittd to th faculty of th Virginia Polytchnic Institut and Stat Univrsity in partial fulfillmnt of th rquirmnts for th dgr of Mastr of Scinc in

More information

Adaptive Load Balancing: A Study in Multi-Agent. Learning. Abstract

Adaptive Load Balancing: A Study in Multi-Agent. Learning. Abstract Journal of Articial Intllignc Rsarch 2 (1995) 475-500 Submittd 10/94; publishd 5/95 Adaptiv Load Balancing: A Study in Multi-Agnt Larning Andra Scharf ascharf@dis.uniroma1.it Dipartimnto di Informatica

More information

Computation and Analysis of Propellant and Levitation Forces of a Maglev System Using FEM Coupled to External Circuit Model

Computation and Analysis of Propellant and Levitation Forces of a Maglev System Using FEM Coupled to External Circuit Model J. Elctromagntic Analysis & Applications, 00, : 70-75 doi:0.436/jmaa.00.4034 Publishd Onlin April 00 (http://www.scirp.org/journal/jmaa) Computation and Analysis of Propllant and Lvitation Forcs of a Maglv

More information

Application of the Topological Optimization Technique to the Stents Cells Design for Angioplasty

Application of the Topological Optimization Technique to the Stents Cells Design for Angioplasty Application of th opological Optimization chniqu to th. A. Guimarãs asptobias@yahoo.com.br S. A. G. Olivira Emritus Mmbr, ABCM sgoulart@mcanica.ufu.br M. A. Duart Snior Mmbr, ABCM mvduart@mcanica.ufu.br

More information

Available online at ScienceDirect. Procedia Materials Science 6 (2014 )

Available online at   ScienceDirect. Procedia Materials Science 6 (2014 ) Availabl onlin at www.scincdirct.com ScincDirct Procdia Matrials Scinc 6 ( ) 6 67 rd Intrnational Confrnc on Matrials Procssing and Charactrisation (ICMPC ) Modal analysis of Functionally Gradd matrial

More information

Artificial Neural Network to the Control of the Parameters of the Heat Treatment Process of Casting

Artificial Neural Network to the Control of the Parameters of the Heat Treatment Process of Casting A R C H I V E S o f F O U N D R Y E N G I N E E R I N G Publishd quartrly as th organ of th Foundry Commission of th Polish Acadmy of Scincs ISSN (1897-3310) Volum 15 Issu 1/2015 119 124 22/1 Artificial

More information

Manufacture of conical springs with elastic medium technology improvement

Manufacture of conical springs with elastic medium technology improvement Journal of Physics: Confrnc Sris PAPER OPE ACCESS Manufactur of conical springs with lastic mdium tchnology improvmnt To cit this articl: S A Kurguov t al 18 J. Phys.: Conf. Sr. 944 169 Viw th articl onlin

More information

RULES REDUCTION AND OPTIMIZATION OF FUZZY LOGIC MEMBERSHIP FUNCTIONS FOR INDUCTION MOTOR SPEED CONTROLLER

RULES REDUCTION AND OPTIMIZATION OF FUZZY LOGIC MEMBERSHIP FUNCTIONS FOR INDUCTION MOTOR SPEED CONTROLLER RULES REDUCTION AND OPTIMIZATION OF FUZZY LOGIC MEMBERSHIP FUNCTIONS FOR INDUCTION MOTOR SPEED CONTROLLER 1 ZULHISYAM SALLEH, 2 MARIZAN SULAIMAN, 3 FIZATUL AINI PATAKOR, 4 ROSLI OMAR 1 Snior Lcturr. Dpt.

More information