Spectral method for deconvolving a density Marine Carrasco Department of Economics University of Rochester Jean-Pierre Flor

Størrelse: px
Begynne med side:

Download "Spectral method for deconvolving a density Marine Carrasco Department of Economics University of Rochester Jean-Pierre Flor"

Transkript

1 Spectral method for decovolvig a desity Marie Carrasco Departmet of coomics Uiversity of Rochester csco@troi.cc.rochester.edu Jea-Pierre Flores GRMAQ ad IDI, Uiversite Toulouse I ores@cict.fr Prelimiary versio: Jauary, 00 Commets Welcome. Itroductio Assume we observe i.i.d. realizatios, x ::: x of the radom variable X with ukow desity h ad X satises X = Y + : where Y ad are idepedet radom variables with probability desity fuctios (p.d.f.) f ad g respectively so that h = f g: The aim of this paper is to give a estimator of f assumig g is kow. This problem cosists i solvig i f the equatio h (x) = f(y)g(x ; y)dy (.) quatio (.) is a itegral equatio ad solvig (.) is typically a ill-posed problem (Tikhoov ad Arsei, 977). Decovolutio is kow to be dicult. Fa (993) gives the optimal speed of covergece for estimators of f ad its derivatives, whatever the method used. The speed of covergece of the Mea Itegrated Square rror (MIS) is particularly slow for supersmooth distributios, for istace it is (l()) ; whe g is the pdf of the ormal distributio ad f is twice dieretiable. The most popular approach to decovolutio is the use of a kerel estimator of f obtaied by applyig the Fourier iversio formula to the empirical characteristic fuctio of X: This method was iitiated by the papers of Carroll ad Hall (988) ad Stefaski ad Carroll (990), later followed We thak James Stock for suggestig this lie of research to us. We are grateful to Susae Scheach for her isightful commets. We also wish to thak the participats of CM coferece (Rochester, 00). This paper is available at

2 by Fa (99a,b, ad 99) amog others. Decovolutio kerel estimators have bee applied i ecoometrics by e.g. Horowitz ad Markatou (996). Recetly, Pesky ad Vidakovic (999) who deal with the estimatio of a decovolutio desity usig a wavelet decompositio. The method we propose here cosists i iterpretig (.) as a itegral equatio Tf = h (.) where T is a compact operator with respect to a well-chose referece space ad therefore admits a coutable iite umber of eigevalues ad eigefuctios. We ivert (.) usig the spectral decompositio of T: We show that if we do oied assumptios o f ad g more precisely if f is smoother tha g the our estimator achieves a much faster rate of covergece tha that obtaied without oied assumptios. I particular, we show that if f ad g are the pdf of two ormal distributios ad the variace of the error (g) is smaller tha that of the sigal (f) the the rate of the MIS is ;= : For a survey o ill-posed problems i the statistical literature ad examples o decovolutio, see Carroll et al. (99) ad Va Rooi ad Ruymgaart (999). They show how to treat the decovolutio problem by solvig the ill-posed problem (.) however they do ot use the trasformatio we use here to make T i (.) compact, therefore they ivert a operator that has a cotiuum of eigevalues. Moreover, they do ot discuss the practical implemetatio of such a estimator. Va Rooi ad Ruymgaart (99) use the same approach to recover a desity i a error-i-variables model o the sphere. The article is orgaized i the followig way. I Sectio, we preset the estimator. I Sectio 3, we establish its speed of covergece. Sectio4 gives its asymptotic ormality. Sectio 5 compares our estimator with the kerel estimator.. Method I Sectio 6, we show that the operator Tf = g(x ; y)f(y)dy (.) is ot a Hilbert Schmidt operator with respect to Lebesgue measure o R ad has a cotiuous spectrum. The mai idea of this paper is to make T compact. To do so, we cosider g (x ; y) as a ctitious probability desity fuctio of Y give X deoted Y X (yx) : We choose more or less arbitrarily a desity X that will play the role of the desity ofx: We costruct a oit desity of(x Y ) that is cosistet with what precedes: (x y) = X (x) g(x ; y): Similarly, we dee m Y as the margial desity of Y m Y (y) = (x y)dx:

3 Now, choose a arbitrary fuctio (ot ecessarily a pdf) Y (y) that satises Assumptio. There exists a costat M > 0such that m Y (y) Y (y) <M for ay y R: This assumptio garatees that L Y (R) ) L m (R) which itself implies T Y L X (R)by the law of iterated expectatios where L Y (R)adL X (R) are deed below ad L m (R) isthespace of square itegrable fuctios with respect to m Y Y : Deote L X (R) the space: L X (R) = (x) such that (x) X (x) dx < : ad L Y (R) thespace: L Y (R) = (y) such that (y) Y (y) dy < : Both the ier product i L X (R) ad i L Y (R) will be deoted h: :i ad both the orm i L X (R) ad i L Y (R) will be deoted k:k without cofusio: We dee the operator T which associates to ay fuctio (y) of L Y (R) a fuctio of L X (R) as: So that quatio (.) ca be rewritte as (T)(x) = Y X [ (Y ) X = x] : Tf = h: (.) Now, we dee the adoit of T which associates to ay fuctio (x) of L X (R) a fuctio of L Y (R): g (x ; y) (T X (x) )(y) = Y (y) (x) dx: For coveiece, we deote XY (xy) = g (x ; y) X (x) Y (y) eve though it is ot a coditioal expectatio. I the special case where m Y T is a coditioal expectatio operator: = Y, the (T )(y) = X Y [ (X) Y = y] : Note that i geeral T6= T. Assumptio. Y domiates f: 3

4 If f is cotiuous with respect to (y) Lebesgue measure o R the it is eough to take Y (y) equivalet to (y): The desity f we wish to estimate has to belog to L Y (R) : Assumptio 3. We have f (y) Y (y)dy < : This coditio may seem restrictive however most of the distributios (the ormal, double-expoetial etc) cosidered i the oparametric literature are square-itegrable. Assumptio 4. We have X (x)g(x ; y) X (x) Y (y) X (x) Y (y)dxdy < : This is a suciet coditio for T to be a Hilbert-Schmidt operator ad therefore to be compact (see Duford ad Schwartz, 963, p. 30, Darolles, Flores ad Reault, 998). As a result of compactess, T has a discrete spectrum. Let 0 = be the eigevalues ad ' 0 0 the oit orthoormal basis of T ad T respectively satisfyig: h i i) T ' (Y ) = (X) 0 h i ii) T (X) = ' (Y ) 0 h i iii) T T ' (Y ) = ' (Y ) 0 h i iv) TT (X) = (X) 0: Note that may becomplex, deotes the complex cougate of : Sice g ad X are give, the eigefuctios are either kow explicitly (see xample below) or ca be estimated by simulatios as precisely as wated (see Sectio 5) so that we ca cosider them as kow. Fially to show the cosistecy, we eed the followig assumptio. h i Assumptio 5. var X (X) (X) < for all 0: Note that a suciet coditio for Assumptio 5 is that the p.d.f. h ad X belog to L that is sup h < ad sup X <. Ideed the variace equals var h X (X) (X) i = X(x) (x)h (x) dx ; X (x) (x)h (x) dx : It is eough to show that the rst term is bouded X(x) (x)h (x) dx (sup h) D (:) (:) X (:) (sup h) X (sup h)(sup X) < : 4

5 quatio (.) ca be approximated by a well-posed problem usig the Tikhoov regularizatio method ( I + T T ) f = T h where the pealizatio term plays the role of the smoothig parameter i the kerel estimatio. Other regularizatio methods could have bee cosidered such as the trucatio which cosists i discardig the smallest eigevalues, however we prefer to use the Tikhoov regularizatio. f becomes f (y) = The oly ukow is T h: Note that D T h ' + ' (y): (.3) (T h)(y) = [h (X) Y = y] A atural estimator of T h is give by dt h (y) = = h (x) XY (xy) dx = h i XY (Xy) X i= So that the estimator of f takes the followig form ^f(y) = X i= XY (x i y) : (.4) D + XY (x i :) ' (:) ' (y): (.5) Remark that f ca be rewritte i the followig fashio: f (y) = = = D h T ' + ' (y) D h + ' (y) h + (X i ) X (X i ) i ' (y): Hece aother expressio of ^f is give by ^f(y) = X i= + (x i ) X (x i ) ' (y): (.6) This expressio requires the estimatio of as well as that of ' however the estimatio of ca be obtaied as a by-product of that of ' without much extra calculatio as explaied i Sectio 5. 5

6 xample. Assume N(0, ): We set x ; y Y X (yx) =g (x ; y) = where deotes the p.d.f. of a stadard ormal. A simple choice for X is x X (x) = so that X a N (0, ) where a meas that we do as if X were distributed as a ormal. The fact that the true distributio may be totally dieret doesot matter. We eed to determie a) Y b) XY that match Y X ad X : a) Y (y) = Y X (yx) X (x) dx = y p : + b) XY (xy) = Y X(yx) X (x) Y (y) = x ; y p where = =( + ): So that we have X Y a N 0 0 : + To calculate the eigevalues ad eigefuctios, we eed to compute T T the expectatio operator associated with (~yy) = Y ~ Y Y X (yx) XY (x~y) dx = ~y ; y p + The eigefuctios of T T are the Hermite polyomials orthoormal with respect to Y associated with the eigevalues : : 6

7 3. Speed of covergece 3.. Rate of the MIS i the geeral case The criterio we use is the MIS with respect to Y that is MIS = ^f ; f = ^f(y) ; f(y) Y (y)dy : The criterio usually employed i the kerel literature (e.g. Stefaski ad Carroll, 990) is the MIS with respect to Lebesgue measure o R. However here ^f(y) is ot ecessarily square-itegrable o R, therefore we replace the itegratio with respect to Lebesgue by a itegratio with respect to Y (:). The rates obtaied by Stefaski ad Carroll are ot aected by the choice of the orm so that their rates will be directly comparable with ours. The MIS ca be rewritte as MIS = = ^f(y) ; f (y)+f (y) ; f(y) Y (y)dy ^f(y) ; f (y) Y (y)dy + Var+ Bias (f (y) ; f(y)) Y (y)dy because ^f = f : As i the kerel estimatio, there is a trade-o betwee the variace (decreasig i ) ad the bias (icreasig i ): Propositio 3.. Uder Assumptios to 5, we have MIS = h i var + X (X i ) (X i ) + Df ' + (3.) A : Df ' This boud depeds o the speed at which the ier products D f ' coverge to zero with : 3.. Automatic selectio of the smoothig parameter The pealizatio term must be selected to miimize the MIS give i (3.). Deote ^f D a estimator of f obtaied usig a ooptimal (quite large) deoted : A estimator of f ' is give by D ^f ' = X i= + X(x i ) (x i ): 7

8 Let ^ ^' ad ^ = ::: B be the estimators of ' ad obtaied by the method h d i described i Sectio 5. Deote var X (X i ) (X i ) the sample variace of X (X i )^(X i ): A estimator of the MIS is give by MIS = 0 BX B + C ^ A = h d i var X (X i ) (X i ) + 0 BX B + C ^ A = P i= X(x i )^ (x i ) o + ^ : (3.) This espressio ca be miimized umerically with respect to to obtai the optimal smoothig parameter Rate i special cases To obtai D further results o the rate of the MIS, we eed D extra assumptios o the ier product f ' : Here, we ivestigate the case where f ' decays at least as fast as : The advatage of this assumptio is to deliver simple results o the rate of covergece of the MIS. It is by o meas the oly assumptio we could have made, we could impose a weaker requiremet that would, D of course, deliver a slower speed of covergece of the MIS. A suciet coditio for f ' to decay at least as fast as is the followig: Coditio A. There exists a fuctio k such that ad kkk = T k = f k (x) X (dx) < : Propositio 3.. Uder Assumptios to 5 ad Coditio A, by choosig a regularizatio parameter = d ;=3 for some d>0 we have MIS = O ;=3 : This rate is a upper boud, we might be able to derive a optimal rate i some special cases. The rate of covergece of the MIS, ;=3 is slower tha ;4=9 obtaied for the kerel estimator i presece of a double-expoetial error ad of course much faster tha [l ()] ; the rate obtaied i presece of a ormal error. Remark. Uder Coditio A, we have Var C Bias = kkk + (3.3) D k + (3.4) 8 + (3.5)

9 So that for = d ;= the rate of covergece of the MIS is give by MIS + : (3.6) Whe oe kows the decay rate of the, oe has immediately the rate of the MIS. Note that the rate of is idepedet of the rate of f: Cosider the case where istead of usig the Tikhoov regularizatio, we had used the trucatio that is we replace by zero ad P by P N where N plays the role of a smoothig parameter. The optimal speed of covergece of N would be dictated by the decay rateof ad therefore depeds o f ad g: Lemma 3.3. If g is eve, that is the error has a symmetric distributio aroud zero ad for X (x) = I [; ] (x) = ad Y (y) = for all y R: A suciet coditio for Coditio A to hold is f (t) g (t) dt < (3.7) where f ad g are the characteristic fuctios of f ad g respectively. Coditio (3.7) requires that f has er tails tha g : Sice the tail behavior of a pdf is related to the smoothess of the pdf, this is equivalet to require that f be smoother tha g (see Ushakov, 999, Theorem.5.4). I the case of f Laplacia, this is a very weak requiremet. I the case of f ormal (discussed i details below), it is less likely to be fullled. Aother iterpretatio of this coditio is the followig: f ca be writte as the covolutio of g ad aother distributio. Va Rooi ad Ruymgaart (99) give a good ituitio o the diculty of decovolutio. If g is smooth the h is also smooth. If f is ot a priori kow to be smooth itself, the problem of recoverig a osmooth f from a sample of smooth h is particularly hard. xample cotiued (ormal case). Cosider ormally distributed ad X ormal ad Y is the margial of (x y). Corollary 3.4. Assume Coditio A holds. = d ;= for some d>0 we have MIS = O By choosig a regularizatio parameter ;= : Result. If N(0 ) ad if f satises exp t f (t) dt < : (3.8) where f (t) is the characteristic fuctio of f the Coditio A is satised. I particular, if Y N(0 v ) ad if v > the Coditio A is satised. 9

10 The proof of Result isgive i Appedix. Remark. Note that i Lemma 3.3, X is bouded ad Y = while, i Result, X is ormal ad Y is the margial of (x y) : Coditio (3.8) requires that the fuctio f be supersmooth, actually it eeds to have er tails tha g: This coditio is atural because it requires i the case of ormal distributios that the variace of the sigal be larger tha the variace of the oise. 4. Asymptotic ormality Because we have iid data, a suciet coditio for asymptotic ormality ^f (y) ; ^f (y) r var ^f (y) L N(0 ) is that the Lyapouov's coditio holds (Billigsley, 995, Theorem 7.3), i.e. for some >0 ; ( ) + 0 (4.) = += [var ( )] where i = D + XY (x i :) ' (:) ' (y): (4.) The coditio (4.) is satised uder the followig assumptios. Assumptio 6. We have X = + 3 0: This coditio requires that go to zero ot too fast. It is satised i the case of a ormal error whe = d ;=, see quatio (8.4). Assumptio 7. There is a costat M idepedet of such that h X (x ) (x ) ; X i 3 <M: Assumptio 7 is trivially satised i most cases. Propositio 4.. Uder Assumptios to 7, if 0 ad,we have ^f (y) ; f (y) r var ^f (y) L N(0 ) : 0

11 The followig assumptio isures that var ^f (y) ca be replaced by the sample variace. Assumptio 8. There is a costat M idepedet of such that h X (x ) (x ) i 4 <M: ad X + 4 0: Lemma 4.. Uder Assumptios -8, we have X i ; ( i ) P 0 i= X i= i ; i P 0: The followig assumptio guaratees that the bias goes to zero sucietly fast so that f ca be replaced by f. Assumptio 9. P D + P f ' 0 + I the case of a ormal error, assumig Coditio A ad = o( ;= ) Assumptio 8 is satised. To guaratee that the bias becomes egligible i compariso with the variace, we have to let go to zero faster tha the optimal rate. Propositio 4.3. Uder Assumptios to 9, if 0 ad,we have p ^f (y) ; f (y) where s = P i= 5. Implemetatio s L N(0 ) i ; P i= i where i is give by (4.). I this sectio, we discuss the practical aspects of the estimatio of f whe o explicit expressio of the eigevalues ad eigefuctios is available. Below, we explai how to estimate the eigevalues ad eigefuctios. Calculatio of eigevalues ad eigefuctios. We are lookig for the solutios of T T' = ': (5.)

12 The coditioal expectatio T ad T could be estimated by kerel but there is a simpler way. a) To estimate the operator T, we will use importace samplig (Geweke, 988). Deote a pdf, such that it is easy to draw data from the distributio correspodig to either by iversio of the c.d.f. or by a reectio method (see Devroye, 986). The operator T ca be estimated by (T')(x) = B = BX b= '(y)g(x ; y)dy '(y)g(x ; y) (y)dy (y) '(y b )g(x ; y b ) : (y b ) where (y b ) b = ::: B is a i.i.d. sample draw from : b) The operator T ca be estimated by (T )(y) = = Y (y) B R BX b= (x) XY (xy) dx (x) X (x) g(x ; y)dx Y (y) (x b ) g(x b ; y): where (x b ) b= ::: B is a i.i.d. sample draw from X : This way we obtai estimators of T ad T that are p B -cosistet ad do ot require a choice of a kerel ad a badwidth. Therefore T T'(y) ca be approximated by Y (y) B BX b= " B BX c= '(y c )g(x b ; y c ) (y c ) # g(x b ; y): This operator has a ite rak ad has at most B eigevalues. Note that the eigefuctios are ecessarily of the form ' (y) = BX b= b g(x b ; y) : (5.) Y (y) Replacig i ' by its expressio, we see that solvig (5.) is equivalet to dig the eigevalues ad eigevectors of the B B;matrix M with pricipal elemet: M b l = B B X c= g(x l ; y c )g(x b ; y c ) : Y (y c )(y c )

13 Let = h Bi 0 be the th eigevector of M associated with, the ' solutio of (5.) is the th eigefuctio of T T associated with the same eigevalue : The fuctio ' ca be evaluated at all poits. Note that the ' associated with distict eigevalues are essarily orthogoal, evertheless, they eed to be omalized. Deote ^' ad ^ the estimators of ' ad. The operator TT (y) cabe approximated by B b= " BX BX c= B B X c= $ (x x c ) (y c ): (y c )g(x c ; y b )g(x ; y b ) (y b ) Y (y b ) # It is easy to verify that the eigefuctios are of the form P B c= c $ (x x c) where = h Bi 0 = ::: are agai the eigevectors of M deed above. Hece the estimators of are give by ^ (x) = BX b= b " BX l= # g(x b ; y l )g(x ; y l ) : (y l ) Y (y l ) Calculatio of ^f: I formula (.5), we eed to compute the term It ca be approximated by D D XY (x i :) ' (:) = d XY (x i :) ' (:) = B XY (x i y)' (y) Y (y)dy: BX b= X (x i ) g(x i ; y b ) ^' (y b ) (y b ): where (y b ) b = ::: B is a i.i.d. sample draw from : Hece we obtai ^f : ^f(y) = X BX i= = + D ^ d XY (x i :) ' (:) ^' (y): We could do a theory that would take ito accout the approximatio error due to the estimatio of the eigevalues ad eigefuctios. But this theory seems useless because B ca be chose arbitrarily large ad we ca make the approximatio error arbitrarily small. 6. Compariso with the kerel desity estimator 6.. Properties of the operator T i L (R ) We rst ivestigate the properties of the operator T deed by (.) if the space of referece is L (R ) : the space of the square-itegrable fuctios with respect to Lebesgue 3

14 measure o R : This correspods to choose X (x) = ad Y (y) = for all x ad y: First ote that the resultig operator T is ot a Hilbert Schmidt operator because R R [g(x ; y)] dxdy is ot bouded. I the followig, we show that T T has a cotiuous spectrum with t = g (t) g (t) for t R where g (t) = R e itz g(z)dz is the characteristic fuctio of the radom variable : Because X ad Y are set equal to oe, we obtai So that We wat to solve Y X (yx) = g(x ; y) XY (xy) = Y X(yx) X (x) Y (y) T [ (x)] = (T T')(y) = (x) g(x ; y)dx: g(x ; y)dx = g(x ; y): g(x ; )'()d = '(y): We take the Fourier trasform o each side: e ity dy g(x ; y)dx g(x ; )'()d = e ity '(y)dy: We apply two successive chages of variables y = x ; u ad x = v + to obtai: g (t) g (t) e it '()d = e ity '(y)dy So that there is a cotiuous spectrum with g (t) g (t) = : t = g (t) '(y) = e ity = p associated with t (x) = e itx = p associated with t : These eigefuctios are othogoormal o [; ] : To check our calculatio, we write f o this basis of eigefuctios: Df ' ' (y) = f (x) ' (x) dx ' (y) = e ;itx f(x)dx e ity dt: We apply the chage of variable u = ;t to obtai: e iux f(x)dx e ;iuy du = f (u) e ;iuy du = f(y) by the Fourier iversio formula. 4

15 6.. Kerel estimator We ca apply i a heuristic maer the same method as before to estimate f:i formula (.5), the elemet DT h ' = Dh ca be estimated by = X i= h (x) (x)dx (x i ): Now, we replace the sum over by a itegral over t ad the eigevalues ad eigefuctios by their expressios i (.5) to obtai ^f(y) = X l= g (t) g (t) + e ;itxl e ity dt: We applyachage of variable u = ;t ^f(y) = X l= g (u) g (u) + e iu(xl;y) du: (6.) Usig the type of regularizatio advocated by Carroll ad al. (99, xample 3..), we get ^f(y) = X f g (u)g g (u) eiu(x l;y) du: l= Now compare these two expressios with the kerel estimator (see e.g. Stefaski ad Carroll, 990). For a smoothig parameter ad a kerel K, the kerel estimator is give by ^f k (y) = X K (u) g (u=) eiu(x l;y)= du: (6.) l= The formulas (6.) ad (6.) dier oly by the way the smoothig is applied. This suggests that the kerel estimator is obtaied by ivertig a operator that has a cotiuous spectrum. Because this spectrum is give by the characteristic fuctio of g, the speed of covergece of the estimator depeds o the behavior of g i the tails. For a formal expositio, see Carroll et al (99, xample 3..). They assume i particular that the fuctio to estimate is p dieretiable ad they obtai a rate of covergece (as a fuctio of p) that is of the same order as the rate of the kerel estimator. 5

16 7. Coclusio It should be emphasized that our approach is fudametally dieret from that of the kerel estimatio. We approximate the fuctio to estimate by a sequece of orthoomal fuctios give by the eigefuctios of the covolutio operator. We show that this estimator is cosistet ad asymptotically ormal. We study its rate of covergece uder extra coditios relatig the smoothess of g with the smoothess of f: We show that uder these extra assumptios, the MIS achieves a fast (arithmetic) rate of covergece. A task that remais to be doe is to prove that, uder geeral assumptios, our estimator achieves the optimal rate of covergece. 8. Appedix Proof of Propositio 3.. We examie successively the terms of variace ad bias. The variace: Usig the expressio of ^f give i (.5), we have ^f(y) ; f (y) = var 4 X D + XY (x i :) ' (:) ' (y) 5 : Because the eigefuctios ' are orthoormal with respect to Y,we have ^f(y) ; f (y) Y (y)dy = + 3 with = var hd i XY (X i :) ' (:) = var XY (X i y) ' (y) Y (y)dy = var X (X i ) Y X (yx i ) ' (y)dy = var h i X (X i ) (X i ) = var h X (X i ) (X i ) i : (8.) So that the variace term ca be maored: Var = A var h i + X (X i ) (X i ) + by Assumptio 4 where A is some costat. 6

17 Bias: Usig (.3), f ca be rewritte as f = = = D h T ' + ' D h + ' D f ' + ' because h = Tf: We have f ; f = I ; ( I + T T ) ; T T = ( I + T T ) ; f = f D f ' + ' It follows that kf ; f k = + : Df ' Proof of Propositio 3.. From the previous sectio, we have Var A for some A>0 (8.) ad assumig Coditio A, we have kf ; f k = = + Dk T' D k + (8.3) Usig the fact that + 4 we obtai kf ; f k 4 Dk = 4 kkk : 7

18 Hece, we obtai a maoratio of the MIS MIS A + 4 kkk : For of order = =3, we have MIS C =3 : Proof of Corollary 3.4. We look for a equivalet of the series i (3.3) ad (3.5). For this, we use the followig result. Let f() be the elemet of a series ad assume f()is a positive ad cotiuous decreasig fuctio of : The it is easy to see that J 0 f (s) ds + f (J) JX f() Whe goes to zero, a equivalet oftheseries is give by f() J 0 0 f (s) ds + f (0) for all J : f (s) ds: I the ormal case, the eigevalues satisfy = with < so that as goes to zero: The rate for the MIS follows. Proof of Lemma l() (8.4) T k = f, g (x ; y) X (x) k (x) dx = Y (y) f (y), g (x ; y) k (x) dx = f (y) : (8.5) where k X k, f Y f: Deote F (g) F (k ) F (f ) the Fourier trasforms of g, k, ad f that is F (g)(t) = R e ;itx g (x) dx: (8.5) is equivalet to F (g) F (k ) = F (f ), k (x) = e F itx (f )(t) F (g)(t) dt, k (x) = e F itx (f Y )(t) X (x) F (g)(t) dt for ay x i the support of X 8

19 by the iversio formula. The coditio R k (x) X (x) dx < is equivalet to Take X =0:5I [; ] ad Y X (x) e F itx (f Y )(t) F (g)(t) dt dx < : (8.6) =: (8.6) is satised as soo as F (f)(t) F (g)(t) dt < : Usigachage of variables t ;t, this is equivalet to Proof of Result. f (t) g (t) dt < : T k = f, x ; y p k (x) dx = f (y), ~ (x ; y) k (x) dx = f (y) where ~ (x) = x= p : Let x = u, we have ~ (u ; y) k (u) du = f (y), (u ; y) k (u) du = f (y) where k (u) = k(u) (v) = (v p =) : Hece we have It follows that k (u) = k (x) = k = f, k = f, k (t) = exp t e ;itu exp t e ;itx= exp t f (t) : f (t) dt f (t) dt: 9

20 The coditio A is equivalet to e ;itx= exp t It is suciet that exp t k (x) exp ; x dx <, f (t) dt exp ; x dx < : f (t) dt < : (8.7) Hece the result. If f is the pdf of N (0 v ) quatio (8.7) is satised i v > = which is satised for a such that > v : (8.8) Sice the rate of covergece of the eigevalues is faster for small we have iterest to choose so that is as small as possible ad satises (8.8). I practice, is assumed to be kow ad v is ukow but ca be easily estimated from the data. Proof of Propositio 4. First we check that var ( ) is bouded from below. var ( )= X X + ' (y)+ <k + + k i ' (y) ' k (y) where D i = cov XY (x i :) ' (:) D XY (x i :) ' k (:) = k cov X X k : usig the same rewritig as i (8.). var ( ) is a sum of positive terms, it is bouded from below. To establish (4.) for =,we eed to show that ; ( ) 3 = 0: D Usig XY (x :) ; T h ' (:) = h X (x ) (x ) ; as We have ; ( ) 3 ; ( )= h i + X ; X ' (y): X i ca be rewritte 3 h i 3 + X ; ' X (y) 3 +cross;products: 0

21 The cross-products are domiated by the rst term. The result follows from Assumptios 6 ad 7. Proof of Lemma4. var ( )=O X A : + Uder Assumptio 8, var ( ) 0 which implies the weak law of large umbers by Theorem C of Serig (980, p. 7). For the WLLN of i, we use 0 var = X 4 A : + Proof of Propositio 4.3. Uder Assumptio 9, we have f ; f var = ^f P D f ' + ' var ( ) coverges to zero. By Assumptio 8 ad Lemma 4., var ( ) ca be replaced by the sample variace. 9. Refereces Billigsley, P. (995) Probability ad Measure, Wiley & Sos, New York. Carroll, R. ad P. Hall (988) \Optimal Rates of Covergece for Decovolvig a Desity", Joural of America Statistical Associatio, 83, No.404, Carroll, R., A. Va Rooi, ad F. Ruymgaart (99) \Theoretical Aspects of Ill-posed Problems i Statistics", Acta Applicadae Mathematicae, 4, Devroye, L. (986) No-Uiform Radom Variate Geeratio, Spriger Verlag, NY. Darolles, S., J.-P. Flores ad. Reault (998), mimeo. Duford, N. ad J. Schwartz (963) Liear Operators, Part II, Spectral Theory, Wiley & Sos, NY. Fa, J. (99a) \O the optimal rates of covergece for oparametric decovolutio problems", The Aals of Statistics, 9, No.3, Fa, J. (99b) \Asymptotic ormality for decovolutio kerel desity estimators", Sakhya, 53, Fa, J. (99) \Decovolutio with supersmooth distributios", The Caadia Joural of Statistics, 0, Fa, J. (993) \Adaptively local oe-dimetioal subproblems with applicatio to a decovolutio problem", The Aals of Statistics,,

22 Geweke, J. (988) \Atithetic Acceleratio of Mote Carlo Itegratio i Bayesia Iferece", Joural of coometrics, 38, Horowitz, J. ad M. Markatou (996) Semiparametric stimatio of Regressio Models for Pael Data, Review of coomic Studies, 63, Pesky, M. ad B. Vidakovic, (999) \Adaptive wavelet estimator for oparametric dedity decovolutio", The Aals of Statistics, 7, Rooi, A. C. M. Va ad F. H. Ruymgaart (99) \Regularized Decovolutio o the Circle ad the Sphere", , i Noparametric Fuctioal stimatio ad Related Topics, d. by G. Roussas, Kluwer Academic Publisher, The Netherlads. Rooi, A. C. M. Va ad F. H. Ruymgaart (999) \O Iverse stimatio", i Asymptotics, Noparametrics ad Time Series, , Dekker, New York. Serflig, R. (980) Approximatio Theorems of Mathematical Statistics, Wiley & Sos, New York. Stefaski, L. ad R. Carroll (990) \Decovolutig Kerel Desity stimators", Statistics,, Tikhoov, A. ad V. Arsei (977) Solutios of Ill-posed Problems, Wisto & Sos, Washigto D.C. Ushakov, N. (999) Selected Topics i Characteristic Fuctios, VSP, Utrecht, The Netherlads.

Continuity. Subtopics

Continuity. Subtopics 0 Cotiuity Chapter 0: Cotiuity Subtopics.0 Itroductio (Revisio). Cotiuity of a Fuctio at a Poit. Discotiuity of a Fuctio. Types of Discotiuity.4 Algebra of Cotiuous Fuctios.5 Cotiuity i a Iterval.6 Cotiuity

Detaljer

MU-Estimation and Smoothing

MU-Estimation and Smoothing Joural of Multivariate Aalysis 76, 277293 (2001) doi:10.1006jmva.2000.1916, available olie at http:www.idealibrary.com o MU-Estimatio Smoothig Z. J. Liu Mississippi State Uiversity C. R. Rao Pe State Uiversity

Detaljer

TMA4245 Statistikk. Øving nummer 12, blokk II Løsningsskisse. Norges teknisk-naturvitenskapelige universitet Institutt for matematiske fag

TMA4245 Statistikk. Øving nummer 12, blokk II Løsningsskisse. Norges teknisk-naturvitenskapelige universitet Institutt for matematiske fag Vår 205 Norges tekisk-aturviteskapelige uiversitet Istitutt for matematiske fag Øvig ummer 2, blokk II Løsigsskisse Oppgave a - β agir biles besiforbruk i liter/mil - Rimelig med α 0 fordi med x 0 ige

Detaljer

Gir vi de resterende 2 oppgavene til én prosess vil alle sitte å vente på de to potensielt tidskrevende prosessene.

Gir vi de resterende 2 oppgavene til én prosess vil alle sitte å vente på de to potensielt tidskrevende prosessene. Figure over viser 5 arbeidsoppgaver som hver tar 0 miutter å utføre av e arbeider. (E oppgave ka ku utføres av é arbeider.) Hver pil i figure betyr at oppgave som blir pekt på ikke ka starte før oppgave

Detaljer

The Accuracy and the Computational Complexity of a Multivariate Binned Kernel Density Estimator 1

The Accuracy and the Computational Complexity of a Multivariate Binned Kernel Density Estimator 1 Joural of Multivariate Aalysis 72, 264309 (2000) Article ID jmva.1999.1863, available olie at httpwww.idealibrary.com o The Accuracy ad the Computatioal Complexity of a Multivariate Bied Kerel Desity Estimator

Detaljer

Second Order Hadamard Differentiability in Statistical Applications

Second Order Hadamard Differentiability in Statistical Applications Joural of Multivariate Aalysis 77, 187228 (21) doi:1.16jmva.2.1926, available olie at http:www.idealibrary.com o Secod Order Hadamard Differetiability i Statistical Applicatios Jia-Jia Re 1 Tulae Uiversity

Detaljer

ASYMPTOTIC NORMALITY OF THE MAXIMUM LIKELIHOOD ESTIMATOR IN. Abstract. State space models is a very general class of time series models capable of

ASYMPTOTIC NORMALITY OF THE MAXIMUM LIKELIHOOD ESTIMATOR IN. Abstract. State space models is a very general class of time series models capable of ASYMPTOTIC NORMALITY OF THE MAIMUM LIKELIHOOD ESTIMATOR IN STATE SPACE MODELS Jes Ledet Jese 1 Niels Vver Peterse 2 Uiversity of Aarhus 1;2 ad MaPhySto 1 Abstract State space models is a very geeral class

Detaljer

On Parameters of Increasing Dimensions

On Parameters of Increasing Dimensions Joural of Multivariate Aalysis 73, 0135 (2000) doi:10.1006jmva.1999.1873, available olie at http:www.idealibrary.com o O Parameters of Icreasig Dimesios Xumig He 1 Departmet of Statistics, Uiversity of

Detaljer

TMA4245 Statistikk Eksamen 20. desember 2012

TMA4245 Statistikk Eksamen 20. desember 2012 Norges tekisk-aturviteskapelige uiversitet Istitutt for matematiske fag TMA4245 Statistikk Eksame 20. desember 202 Løsigsskisse Oppgave a Sasylighete for å få 5 kro er P 5 kro = = /32 = 0.03. 25 Sasylighete

Detaljer

Slope-Intercept Formula

Slope-Intercept Formula LESSON 7 Slope Intercept Formula LESSON 7 Slope-Intercept Formula Here are two new words that describe lines slope and intercept. The slope is given by m (a mountain has slope and starts with m), and intercept

Detaljer

TMA4240 Statistikk Høst 2015

TMA4240 Statistikk Høst 2015 Høst 205 Norges tekisk-aturviteskapelige uiversitet Istitutt for matematiske fag Øvig ummer 2, blokk II Løsigsskisse Oppgave a - β agir biles besiforbruk i liter/mil - Rimelig med α 0 fordi med x 0 ige

Detaljer

Empirical Likelihood Ratio in Terms of Cumulative Hazard Function for Censored Data

Empirical Likelihood Ratio in Terms of Cumulative Hazard Function for Censored Data Joural of Multivariate Aalysis 80, 6688 (2002) doi0.006jmva.2000.977, available olie at httpwww.idealibrary.com o Empirical Likelihood Ratio i Terms of Cumulative Hazard Fuctio for Cesored Data Xiao-Rog

Detaljer

Z. D. Bai. and. Y. Wu. Department of Mathematics and Statistics, York University, 4700 Keele Street, North York, Ontario, Canada M3J 1P3

Z. D. Bai. and. Y. Wu. Department of Mathematics and Statistics, York University, 4700 Keele Street, North York, Ontario, Canada M3J 1P3 joural of multivariate aalysis 63, 119135 (1997) article o. MV971694 Geeral M-Estimatio Z. D. Bai Departmet of Applied Mathematics, Natioal Su Yat-se Uiversity, Kaohsiug, Taiwa ad Y. Wu Departmet of Mathematics

Detaljer

Prediction from Randomly Right Censored Data 1

Prediction from Randomly Right Censored Data 1 Joural of Multivariate Aalysis 80, 7300 (2002) doi:0.006jmva.2000.973, available olie at http:www.idealibrary.com o Predictio from Radomly Right Cesored Data Michael Kohler ad Kiga Ma the Uiversita t Stuttgart,

Detaljer

Asymptotics for Homogeneity Tests Based on a Multivariate Random Effects Proportional Hazards Model

Asymptotics for Homogeneity Tests Based on a Multivariate Random Effects Proportional Hazards Model Joural of Multivariate Aalysis 78, 8310 (001) doi10.1006jmva.000.1940, available olie at httpwww.idealibrary.com o Asymptotics for Homogeeity Tests Based o a Multivariate Radom Effects Proportioal Hazards

Detaljer

Normal Approximation Rate and Bias Reduction for Data-Driven Kernel Smoothing Estimator in a Semiparametric Regression Model

Normal Approximation Rate and Bias Reduction for Data-Driven Kernel Smoothing Estimator in a Semiparametric Regression Model Joural of Multivariate Aalysis 80, 120 (2002) doi:10.1006mva.2000.1925, available olie at http:www.idealibrary.com o Normal Approximatio Rate ad Bias Reductio for Data-Drive Kerel Smoothig Estimator i

Detaljer

Trigonometric Substitution

Trigonometric Substitution Trigonometric Substitution Alvin Lin Calculus II: August 06 - December 06 Trigonometric Substitution sin 4 (x) cos (x) dx When you have a product of sin and cos of different powers, you have three different

Detaljer

Nonparametric analysis of covariance Holger Dette Ruhr-Universitat Bochum Fakultat fur Mathematik Bochum Germany

Nonparametric analysis of covariance Holger Dette Ruhr-Universitat Bochum Fakultat fur Mathematik Bochum Germany oparametric aalysis of covariace Holger Dette Rur-Uiversitat Bocum Faultat fur Matemati 4478 Bocum Germay email: olger.dette@rur-ui-bocum.de FAX: +49 34 7 94 559 atalie eumeyer Rur-Uiversitat Bocum Faultat

Detaljer

Unit Relational Algebra 1 1. Relational Algebra 1. Unit 3.3

Unit Relational Algebra 1 1. Relational Algebra 1. Unit 3.3 Relational Algebra 1 Unit 3.3 Unit 3.3 - Relational Algebra 1 1 Relational Algebra Relational Algebra is : the formal description of how a relational database operates the mathematics which underpin SQL

Detaljer

Average Regression Surface for Dependent Data

Average Regression Surface for Dependent Data Joural of Multivariate Aalysis 75, 242 (2000) doi:0.006jmva.999.896, available olie at http:www.idealibrary.com o Average Regressio Surface for Depedet Data Zogwu Cai Uiversity of North Carolia E-mail:

Detaljer

Conditional Empirical Processes Defined by Nonstationary Absolutely Regular Sequences

Conditional Empirical Processes Defined by Nonstationary Absolutely Regular Sequences Joural of Multivariate Aalysis 70, 250285 (1999) Article ID jmva.1999.1822, available olie at httpwww.idealibrary.com o Coditioal Empirical Processes Defied by Nostatioary Absolutely Regular Sequeces Michel

Detaljer

Emneevaluering GEOV272 V17

Emneevaluering GEOV272 V17 Emneevaluering GEOV272 V17 Studentenes evaluering av kurset Svarprosent: 36 % (5 av 14 studenter) Hvilket semester er du på? Hva er ditt kjønn? Er du...? Er du...? - Annet PhD Candidate Samsvaret mellom

Detaljer

SVM and Complementary Slackness

SVM and Complementary Slackness SVM and Complementary Slackness David Rosenberg New York University February 21, 2017 David Rosenberg (New York University) DS-GA 1003 February 21, 2017 1 / 20 SVM Review: Primal and Dual Formulations

Detaljer

Eksamensoppgave i TMA4320 Introduksjon til vitenskapelige beregninger

Eksamensoppgave i TMA4320 Introduksjon til vitenskapelige beregninger Istitutt for matematiske fag Eksamesoppgave i TMA432 Itroduksjo til viteskapelige beregiger Faglig kotakt uder eksame: Ato Evgrafov Tlf: 453 163 Eksamesdato: 6. jui 216 Eksamestid (fra til): 9: 13: Hjelpemiddelkode/Tillatte

Detaljer

Comparison of LR, Score, and Wald Tests in a Non-IID Setting

Comparison of LR, Score, and Wald Tests in a Non-IID Setting oural of multivariate aalysis 60, 99110 (1997) article o. MV961645 Compariso of LR, Score, ad Wald Tests i a No-IID Settig C. Radhakrisha Rao The Pesylvaia State Uiversity ad Rahul Mukeree Idia Istitute

Detaljer

Dynamic Programming Longest Common Subsequence. Class 27

Dynamic Programming Longest Common Subsequence. Class 27 Dynamic Programming Longest Common Subsequence Class 27 Protein a protein is a complex molecule composed of long single-strand chains of amino acid molecules there are 20 amino acids that make up proteins

Detaljer

A New Approach to the BHEP Tests for Multivariate Normality

A New Approach to the BHEP Tests for Multivariate Normality joural of multivariate aalysis 6, 13 (1997) article o. MV971684 A New Approach to the BHEP Tests for Multivariate Normality Norbert Heze ad Thorste Wager Uiversita t Karlsruhe, D-7618 Karlsruhe, Germay

Detaljer

Permutation Tests for Reflected Symmetry

Permutation Tests for Reflected Symmetry joural of multivariate aalysis 67, 129153 (1998) article o. MV971697 Permutatio Tests for Reflected Symmetry Georg Neuhaus ad Li-Xig Zhu* Uiversity of Hamburg, Hamburg, Germay ad Chiese Academy of Scieces,

Detaljer

UNIVERSITETET I OSLO

UNIVERSITETET I OSLO UNIVERSITETET I OSLO Det matematisk-naturvitenskapelige fakultet Eksamen i MAT2400 Analyse 1. Eksamensdag: Onsdag 15. juni 2011. Tid for eksamen: 09.00 13.00 Oppgavesettet er på 6 sider. Vedlegg: Tillatte

Detaljer

Analysis of Algorithms

Analysis of Algorithms Aalysis of Algorithms Aalysis Fram Most algorithms trasform iput objects ito output objects. The ruig time of a algorithm typically grows with the iput size. Average case time is ofte difficult to determie.

Detaljer

Den som gjør godt, er av Gud (Multilingual Edition)

Den som gjør godt, er av Gud (Multilingual Edition) Den som gjør godt, er av Gud (Multilingual Edition) Arne Jordly Click here if your download doesn"t start automatically Den som gjør godt, er av Gud (Multilingual Edition) Arne Jordly Den som gjør godt,

Detaljer

Physical origin of the Gouy phase shift by Simin Feng, Herbert G. Winful Opt. Lett. 26, (2001)

Physical origin of the Gouy phase shift by Simin Feng, Herbert G. Winful Opt. Lett. 26, (2001) by Simin Feng, Herbert G. Winful Opt. Lett. 26, 485-487 (2001) http://smos.sogang.ac.r April 18, 2014 Introduction What is the Gouy phase shift? For Gaussian beam or TEM 00 mode, ( w 0 r 2 E(r, z) = E

Detaljer

Characteristic Functions of L 1 -Spherical and L 1 -Norm Symmetric Distributions and Their Applications

Characteristic Functions of L 1 -Spherical and L 1 -Norm Symmetric Distributions and Their Applications Joural of Multivariate Aalysis 76, 192213 (2001) doi10.1006jmva.2001.1910, available olie at httpwww.idealibrary.com o Characteristic Fuctios of L 1 -Spherical ad L 1 -Norm Symmetric Distributios ad Their

Detaljer

UNIVERSITY OF OSLO DEPARTMENT OF ECONOMICS

UNIVERSITY OF OSLO DEPARTMENT OF ECONOMICS UNIVERSITY OF OSLO DEPARTMENT OF ECONOMICS Postponed exam: ECON420 Mathematics 2: Calculus and linear algebra Date of exam: Tuesday, June 8, 203 Time for exam: 09:00 a.m. 2:00 noon The problem set covers

Detaljer

Endelig ikke-røyker for Kvinner! (Norwegian Edition)

Endelig ikke-røyker for Kvinner! (Norwegian Edition) Endelig ikke-røyker for Kvinner! (Norwegian Edition) Allen Carr Click here if your download doesn"t start automatically Endelig ikke-røyker for Kvinner! (Norwegian Edition) Allen Carr Endelig ikke-røyker

Detaljer

0.5 (6x 6x2 ) dx = [3x 2 2x 3 ] 0.9. n n. = n. ln x i + (β 1) i=1. n i=1

0.5 (6x 6x2 ) dx = [3x 2 2x 3 ] 0.9. n n. = n. ln x i + (β 1) i=1. n i=1 Norges tekisk-aturviteskapelige uiversitet Istitutt for matematiske fag Øvig ummer 9, blokk II Løsigsskisse Oppgave a The probability is.9.5 6x( x dx.9.5 (6x 6x dx [3x x 3 ].9.5.47. b The likelihood fuctio

Detaljer

Databases 1. Extended Relational Algebra

Databases 1. Extended Relational Algebra Databases 1 Extended Relational Algebra Relational Algebra What is an Algebra? Mathematical system consisting of: Operands --- variables or values from which new values can be constructed. Operators ---

Detaljer

Exercise 1: Phase Splitter DC Operation

Exercise 1: Phase Splitter DC Operation Exercise 1: DC Operation When you have completed this exercise, you will be able to measure dc operating voltages and currents by using a typical transistor phase splitter circuit. You will verify your

Detaljer

Hvor mye teoretisk kunnskap har du tilegnet deg på dette emnet? (1 = ingen, 5 = mye)

Hvor mye teoretisk kunnskap har du tilegnet deg på dette emnet? (1 = ingen, 5 = mye) INF283, HØST 16 Er du? Er du? - Annet Hvor mye teoretisk kunnskap har du tilegnet deg på dette emnet? (1 = ingen, 5 = mye) Hvor mye praktisk kunnskap har du tilegnet deg på dette emnet? (1 = ingen, 5 =

Detaljer

Mathematics 114Q Integration Practice Problems SOLUTIONS. = 1 8 (x2 +5x) 8 + C. [u = x 2 +5x] = 1 11 (3 x)11 + C. [u =3 x] = 2 (7x + 9)3/2

Mathematics 114Q Integration Practice Problems SOLUTIONS. = 1 8 (x2 +5x) 8 + C. [u = x 2 +5x] = 1 11 (3 x)11 + C. [u =3 x] = 2 (7x + 9)3/2 Mathematics 4Q Name: SOLUTIONS. (x + 5)(x +5x) 7 8 (x +5x) 8 + C [u x +5x]. (3 x) (3 x) + C [u 3 x] 3. 7x +9 (7x + 9)3/ [u 7x + 9] 4. x 3 ( + x 4 ) /3 3 8 ( + x4 ) /3 + C [u + x 4 ] 5. e 5x+ 5 e5x+ + C

Detaljer

Hvor mye praktisk kunnskap har du tilegnet deg på dette emnet? (1 = ingen, 5 = mye)

Hvor mye praktisk kunnskap har du tilegnet deg på dette emnet? (1 = ingen, 5 = mye) INF247 Er du? Er du? - Annet Ph.D. Student Hvor mye teoretisk kunnskap har du tilegnet deg på dette emnet? (1 = ingen, 5 = mye) Hvor mye praktisk kunnskap har du tilegnet deg på dette emnet? (1 = ingen,

Detaljer

Universitetet i Bergen Det matematisk-naturvitenskapelige fakultet Eksamen i emnet Mat131 - Differensiallikningar I Onsdag 25. mai 2016, kl.

Universitetet i Bergen Det matematisk-naturvitenskapelige fakultet Eksamen i emnet Mat131 - Differensiallikningar I Onsdag 25. mai 2016, kl. 1 MAT131 Bokmål Universitetet i Bergen Det matematisk-naturvitenskapelige fakultet Eksamen i emnet Mat131 - Differensiallikningar I Onsdag 25. mai 2016, kl. 09-14 Oppgavesettet er 4 oppgaver fordelt på

Detaljer

Call function of two parameters

Call function of two parameters Call function of two parameters APPLYUSER USER x fµ 1 x 2 eµ x 1 x 2 distinct e 1 0 0 v 1 1 1 e 2 1 1 v 2 2 2 2 e x 1 v 1 x 2 v 2 v APPLY f e 1 e 2 0 v 2 0 µ Evaluating function application The math demands

Detaljer

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT UNIVERSITETET I OSLO ØKONOMISK INSTITUTT Utsatt ksamen i: ECON3120/4120 Matematikk 2: Matematisk analyse og lineær algebra Postponed exam: ECON3120/4120 Mathematics 2: Calculus and linear algebra Eksamensdag:

Detaljer

5 E Lesson: Solving Monohybrid Punnett Squares with Coding

5 E Lesson: Solving Monohybrid Punnett Squares with Coding 5 E Lesson: Solving Monohybrid Punnett Squares with Coding Genetics Fill in the Brown colour Blank Options Hair texture A field of biology that studies heredity, or the passing of traits from parents to

Detaljer

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT UNIVERSITETET I OSLO ØKONOMISK INSTITUTT BOKMÅL Eksamen i: ECON1210 - Forbruker, bedrift og marked Eksamensdag: 26.11.2013 Sensur kunngjøres: 18.12.2013 Tid for eksamen: kl. 14:30-17:30 Oppgavesettet er

Detaljer

Solutions #12 ( M. y 3 + cos(x) ) dx + ( sin(y) + z 2) dy + xdz = 3π 4. The surface M is parametrized by σ : [0, 1] [0, 2π] R 3 with.

Solutions #12 ( M. y 3 + cos(x) ) dx + ( sin(y) + z 2) dy + xdz = 3π 4. The surface M is parametrized by σ : [0, 1] [0, 2π] R 3 with. Solutions #1 1. a Show that the path γ : [, π] R 3 defined by γt : cost ı sint j sint k lies on the surface z xy. b valuate y 3 cosx dx siny z dy xdz where is the closed curve parametrized by γ. Solution.

Detaljer

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT UNIVERSITETET I OSLO ØKONOMISK INSTITUTT Eksamen i: ECON30/40 Matematikk : Matematisk analyse og lineær algebra Exam: ECON30/40 Mathematics : Calculus and Linear Algebra Eksamensdag: Tirsdag 0. desember

Detaljer

Hvor mye teoretisk kunnskap har du tilegnet deg på dette emnet? (1 = ingen, 5 = mye)

Hvor mye teoretisk kunnskap har du tilegnet deg på dette emnet? (1 = ingen, 5 = mye) INF234 Er du? Er du? - Annet Hvor mye teoretisk kunnskap har du tilegnet deg på dette emnet? (1 = ingen, 5 = mye) Hvor mye praktisk kunnskap har du tilegnet deg på dette emnet? (1 = ingen, 5 = mye) Hvor

Detaljer

Strassen's LIL for the Lorenz Curve

Strassen's LIL for the Lorenz Curve joural of multivariate aalysis 59, 112 (1996) article o. 5 Strasse's LIL for the Lorez Curve Miklo s Cso rgo * ad Ric$ ardas Zitikis - Carleto Uiversity, Ottawa, Otario, Caada We prove Strasse's law of

Detaljer

UNIVERSITETET I OSLO

UNIVERSITETET I OSLO UNIVERSITETET I OSLO Det matematisk-naturvitenskapelige fakultet Eksamen i INF 3230 Formell modellering og analyse av kommuniserende systemer Eksamensdag: 4. juni 2010 Tid for eksamen: 9.00 12.00 Oppgavesettet

Detaljer

0:7 0:2 0:1 0:3 0:5 0:2 0:1 0:4 0:5 P = 0:56 0:28 0:16 0:38 0:39 0:23

0:7 0:2 0:1 0:3 0:5 0:2 0:1 0:4 0:5 P = 0:56 0:28 0:16 0:38 0:39 0:23 UTKAST ENGLISH VERSION EKSAMEN I: MOT100A STOKASTISKE PROSESSER VARIGHET: 4 TIMER DATO: 16. februar 2006 TILLATTE HJELPEMIDLER: Kalkulator; Tabeller og formler i statistikk (Tapir forlag): Rottman: Matematisk

Detaljer

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT UNIVERSITETET I OSLO ØKONOMISK INSTITUTT Eksamen i: ECON360/460 Samfunnsøkonomisk lønnsomhet og økonomisk politikk Exam: ECON360/460 - Resource allocation and economic policy Eksamensdag: Fredag 2. november

Detaljer

Andrew Gendreau, Olga Rosenbaum, Anthony Taylor, Kenneth Wong, Karl Dusen

Andrew Gendreau, Olga Rosenbaum, Anthony Taylor, Kenneth Wong, Karl Dusen Andrew Gendreau, Olga Rosenbaum, Anthony Taylor, Kenneth Wong, Karl Dusen The Process Goal Definition Data Collection Data Preprocessing EDA Choice of Variables Choice of Method(s) Performance Evaluation

Detaljer

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT Eksame i: ECON30 Statistikk Exam: ECON30 Statistics UNIVERSITETET I OSLO ØKONOMISK INSTITUTT Eksamesdag: Tirsdag. jui 00 Sesur kugjøres: Tirsdag 5. jui, ca. 6.00 Date of exam: Tuesday, Jue, 00 Grades will

Detaljer

Neural Network. Sensors Sorter

Neural Network. Sensors Sorter CSC 302 1.5 Neural Networks Simple Neural Nets for Pattern Recognition 1 Apple-Banana Sorter Neural Network Sensors Sorter Apples Bananas 2 Prototype Vectors Measurement vector p = [shape, texture, weight]

Detaljer

Kurskategori 2: Læring og undervisning i et IKT-miljø. vår

Kurskategori 2: Læring og undervisning i et IKT-miljø. vår Kurskategori 2: Læring og undervisning i et IKT-miljø vår Kurs i denne kategorien skal gi pedagogisk og didaktisk kompetanse for å arbeide kritisk og konstruktivt med IKT-baserte, spesielt nettbaserte,

Detaljer

Stationary Phase Monte Carlo Methods

Stationary Phase Monte Carlo Methods Stationary Phase Monte Carlo Methods Daniel Doro Ferrante G. S. Guralnik, J. D. Doll and D. Sabo HET Physics Dept, Brown University, USA. danieldf@het.brown.edu www.het.brown.edu Introduction: Motivations

Detaljer

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT UNIVERSITETET I OSLO ØKONOMISK INSTITUTT Utsatt eksamen i: ECON1410 - Internasjonal økonomi Exam: ECON1410 - International economics Eksamensdag: 18.06.2013 Date of exam: 18.06.2013 Tid for eksamen: kl.

Detaljer

Han Ola of Han Per: A Norwegian-American Comic Strip/En Norsk-amerikansk tegneserie (Skrifter. Serie B, LXIX)

Han Ola of Han Per: A Norwegian-American Comic Strip/En Norsk-amerikansk tegneserie (Skrifter. Serie B, LXIX) Han Ola of Han Per: A Norwegian-American Comic Strip/En Norsk-amerikansk tegneserie (Skrifter. Serie B, LXIX) Peter J. Rosendahl Click here if your download doesn"t start automatically Han Ola of Han Per:

Detaljer

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT UNIVERSITETET I OSLO ØKONOMISK INSTITUTT Eksamen i: ECON1910 Poverty and distribution in developing countries Exam: ECON1910 Poverty and distribution in developing countries Eksamensdag: 1. juni 2011 Sensur

Detaljer

Generalization of age-structured models in theory and practice

Generalization of age-structured models in theory and practice Generalization of age-structured models in theory and practice Stein Ivar Steinshamn, stein.steinshamn@snf.no 25.10.11 www.snf.no Outline How age-structured models can be generalized. What this generalization

Detaljer

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT UNIVERSITETET I OSLO ØKONOMISK INSTITUTT Eksamen i: ECON3120/4120 Matematikk 2: Matematisk analyse og lineær algebra Exam: ECON3120/4120 Mathematics 2: Calculus and Linear Algebra Eksamensdag: Tirsdag

Detaljer

Hvor mye teoretisk kunnskap har du tilegnet deg på dette emnet? (1 = ingen, 5 = mye)

Hvor mye teoretisk kunnskap har du tilegnet deg på dette emnet? (1 = ingen, 5 = mye) Emneevaluering GEOV325 Vår 2016 Kommentarer til GEOV325 VÅR 2016 (emneansvarlig) Forelesingsrommet inneholdt ikke gode nok muligheter for å kunne skrive på tavle og samtidig ha mulighet for bruk av power

Detaljer

Hvor finner vi flått på vårbeiter? - og betydning av gjengroing for flåttangrep på lam på vårbeite

Hvor finner vi flått på vårbeiter? - og betydning av gjengroing for flåttangrep på lam på vårbeite Hvor finner vi flått på vårbeiter? - og betydning av gjengroing for flåttangrep på lam på vårbeite Lucy Gilbert, Lise Grove, Unni Støbet Lande, Ingeborg Klingen, Kirstyn Brunker Gjenngroing På verdensbasis

Detaljer

Optimal Spherical Deconvolution 1

Optimal Spherical Deconvolution 1 Joura o Mutivariate Aaysis doi10.1006jmva.2000.1968, avaiabe oie at httpwww.ideaibrary.com o Optima Spherica Decovoutio 1 Peter T. Kim Uiversity o Gueph, Gueph, Otario, Caada ad Ja-Yog Koo Haym Uiversity,

Detaljer

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT UNIVERSITETET I OSLO ØKONOMISK INSTITUTT Eksamen i: ECON20/420 Matematikk 2: Matematisk analyse og lineær algebra Exam: ECON20/420 Mathematics 2: Calculus and Linear Algebra Eksamensdag: Fredag 2. mai

Detaljer

Information search for the research protocol in IIC/IID

Information search for the research protocol in IIC/IID Information search for the research protocol in IIC/IID 1 Medical Library, 2013 Library services for students working with the research protocol and thesis (hovedoppgaven) Open library courses: http://www.ntnu.no/ub/fagside/medisin/medbiblkurs

Detaljer

Varieties of two-dimensional cylindric algebras II

Varieties of two-dimensional cylindric algebras II Algebra uivers. 51 (004) 177 06 000-540/04/030177 30 DOI 10.1007/s0001-004-1856- c Birkhäuser Verlag, Basel, 004 Algebra Uiversalis Varieties of two-dimesioal cylidric algebras II Nick Bezhaishvili Abstract.

Detaljer

Dagens tema: Eksempel Klisjéer (mønstre) Tommelfingerregler

Dagens tema: Eksempel Klisjéer (mønstre) Tommelfingerregler UNIVERSITETET I OSLO INF1300 Introduksjon til databaser Dagens tema: Eksempel Klisjéer (mønstre) Tommelfingerregler Institutt for informatikk Dumitru Roman 1 Eksempel (1) 1. The system shall give an overview

Detaljer

FIRST LEGO League. Härnösand 2012

FIRST LEGO League. Härnösand 2012 FIRST LEGO League Härnösand 2012 Presentasjon av laget IES Dragons Vi kommer fra Härnosänd Snittalderen på våre deltakere er 11 år Laget består av 4 jenter og 4 gutter. Vi representerer IES i Sundsvall

Detaljer

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT UNIVERSITETET I OSLO ØKONOMISK INSTITUTT Eksamen i: ECON320/420 Matematikk 2: Matematisk analyse og lineær algebra Exam: ECON320/420 Mathematics 2: Calculus and Linear Algebra Eksamensdag: Tirsdag 7. juni

Detaljer

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT UNIVERSITETET I OSLO ØKONOMISK INSTITUTT Eksamen i: ECON320/420 Matematikk 2: Matematisk analyse og lineær algebra Exam: ECON320/420 Mathematics 2: Calculus and Linear Algebra Eksamensdag: Mandag 8. desember

Detaljer

Vekeplan 4. Trinn. Måndag Tysdag Onsdag Torsdag Fredag AB CD AB CD AB CD AB CD AB CD. Norsk Matte Symjing Ute Norsk Matte M&H Norsk

Vekeplan 4. Trinn. Måndag Tysdag Onsdag Torsdag Fredag AB CD AB CD AB CD AB CD AB CD. Norsk Matte Symjing Ute Norsk Matte M&H Norsk Vekeplan 4. Trinn Veke 39 40 Namn: Måndag Tysdag Onsdag Torsdag Fredag AB CD AB CD AB CD AB CD AB CD Norsk Engelsk M& Mitt val Engelsk Matte Norsk Matte felles Engelsk M& Mitt val Engelsk Norsk M& Matte

Detaljer

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT UNIVERSITETET I OSLO ØKONOMISK INSTITUTT Eksamen i: ECON20 Forbruker, bedrift og marked, høsten 2004 Exam: ECON20 - Consumer behavior, firm behavior and markets, autumn 2004 Eksamensdag: Onsdag 24. november

Detaljer

Appendix B, not for publication, with screenshots for Fairness and family background

Appendix B, not for publication, with screenshots for Fairness and family background Appendix B, not for publication, with screenshots for Fairness and family background Ingvild Almås Alexander W. Cappelen Kjell G. Salvanes Erik Ø. Sørensen Bertil Tungodden This document shows screenshots

Detaljer

Speed Racer Theme. Theme Music: Cartoon: Charles Schultz / Jef Mallett Peanuts / Frazz. September 9, 2011 Physics 131 Prof. E. F.

Speed Racer Theme. Theme Music: Cartoon: Charles Schultz / Jef Mallett Peanuts / Frazz. September 9, 2011 Physics 131 Prof. E. F. September 9, 2011 Physics 131 Prof. E. F. Redish Theme Music: Speed Racer Theme Cartoon: Charles Schultz / Jef Mallett Peanuts / Frazz 1 Reading questions Are the lines on the spatial graphs representing

Detaljer

Graphs similar to strongly regular graphs

Graphs similar to strongly regular graphs Joint work with Martin Ma aj 5th June 2014 Degree/diameter problem Denition The degree/diameter problem is the problem of nding the largest possible graph with given diameter d and given maximum degree

Detaljer

2A September 23, 2005 SPECIAL SECTION TO IN BUSINESS LAS VEGAS

2A September 23, 2005 SPECIAL SECTION TO IN BUSINESS LAS VEGAS 2A September 23, 2005 SPECIAL SECTION TO IN BUSINESS LAS VEGAS SPECIAL SECTION TO IN BUSINESS LAS VEGAS 3A September 23, 2005 SEE, PAGE 8A Businesses seek flexibility. It helps them compete in a fast-paced,

Detaljer

PATIENCE TÅLMODIGHET. Is the ability to wait for something. Det trenger vi når vi må vente på noe

PATIENCE TÅLMODIGHET. Is the ability to wait for something. Det trenger vi når vi må vente på noe CARING OMSORG Is when we show that we care about others by our actions or our words Det er når vi viser at vi bryr oss om andre med det vi sier eller gjør PATIENCE TÅLMODIGHET Is the ability to wait for

Detaljer

KROPPEN LEDER STRØM. Sett en finger på hvert av kontaktpunktene på modellen. Da får du et lydsignal.

KROPPEN LEDER STRØM. Sett en finger på hvert av kontaktpunktene på modellen. Da får du et lydsignal. KROPPEN LEDER STRØM Sett en finger på hvert av kontaktpunktene på modellen. Da får du et lydsignal. Hva forteller dette signalet? Gå flere sammen. Ta hverandre i hendene, og la de to ytterste personene

Detaljer

GEO231 Teorier om migrasjon og utvikling

GEO231 Teorier om migrasjon og utvikling U N I V E R S I T E T E T I B E R G E N Institutt for geografi Emnerapport høsten 2013: GEO231 Teorier om migrasjon og utvikling Innhold: 1. Informasjon om emnet 2. Statistikk 3. Egenevaluering 4. Studentevaluering

Detaljer

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT UNIVERSITETET I OSLO ØKONOMISK INSTITUTT Exam: ECON320/420 Mathematics 2: Calculus and Linear Algebra Eksamen i: ECON320/420 Matematikk 2: Matematisk analyse og lineær algebra Date of exam: Friday, May

Detaljer

Hvordan kvalitetssikre åpne tidsskrift?

Hvordan kvalitetssikre åpne tidsskrift? Hvordan kvalitetssikre åpne tidsskrift? Innlegg ved Halvdagsseminar om åpen tilgang til vitenskapelige artikler I forbindelse med den internasjonale Open Access-uken, Universitetsbiblioteket i Bergen 26.

Detaljer

The regulation requires that everyone at NTNU shall have fire drills and fire prevention courses.

The regulation requires that everyone at NTNU shall have fire drills and fire prevention courses. 1 The law The regulation requires that everyone at NTNU shall have fire drills and fire prevention courses. 2. 3 Make your self familiar with: Evacuation routes Manual fire alarms Location of fire extinguishers

Detaljer

Gradient. Masahiro Yamamoto. last update on February 29, 2012 (1) (2) (3) (4) (5)

Gradient. Masahiro Yamamoto. last update on February 29, 2012 (1) (2) (3) (4) (5) Gradient Masahiro Yamamoto last update on February 9, 0 definition of grad The gradient of the scalar function φr) is defined by gradφ = φr) = i φ x + j φ y + k φ ) φ= φ=0 ) ) 3) 4) 5) uphill contour downhill

Detaljer

Splitting the differential Riccati equation

Splitting the differential Riccati equation Splitting the differential Riccati equation Tony Stillfjord Numerical Analysis, Lund University Joint work with Eskil Hansen Innsbruck Okt 15, 2014 Outline Splitting methods for evolution equations The

Detaljer

SAMPOL115 Emneevaluering høsten 2014

SAMPOL115 Emneevaluering høsten 2014 SAMPOL115 Emneevaluering høsten 2014 Om emnet SAMPOL 270 ble avholdt for førsten gang høsten 2013. Det erstatter til dels SAMPOL217 som sist ble avholdt høsten 2012. Denne høsten 2014 var Michael Alvarez

Detaljer

Emnedesign for læring: Et systemperspektiv

Emnedesign for læring: Et systemperspektiv 1 Emnedesign for læring: Et systemperspektiv v. professor, dr. philos. Vidar Gynnild Om du ønsker, kan du sette inn navn, tittel på foredraget, o.l. her. 2 In its briefest form, the paradigm that has governed

Detaljer

EKSAMENSOPPGAVE I BI2034 Samfunnsøkologi EXAMINATION IN: BI Community ecology

EKSAMENSOPPGAVE I BI2034 Samfunnsøkologi EXAMINATION IN: BI Community ecology Norges teknisk-naturvitenskapelige universitet Institutt for Biologi EKSAMENSOPPGAVE I BI2034 Samfunnsøkologi EXAMINATION IN: BI2034 - Community ecology - Faglig kontakt under eksamen/contact person/subject

Detaljer

ECON3120/4120 Mathematics 2, spring 2004 Problem solutions for the seminar on 5 May Old exam problems

ECON3120/4120 Mathematics 2, spring 2004 Problem solutions for the seminar on 5 May Old exam problems Department of Economics May 004 Arne Strøm ECON0/40 Mathematics, spring 004 Problem solutions for the seminar on 5 May 004 (For practical reasons (read laziness, most of the solutions this time are in

Detaljer

Løsningsførslag i Matematikk 4D, 4N, 4M

Løsningsførslag i Matematikk 4D, 4N, 4M Norges teknisk naturvitenskapelige universitet Institutt for matematiske fag Side av 6 Løsningsførslag i Matematikk 4D, 4N, 4M Oppgave (Kun før 4D Vi har f(x, y x + y x y, for x y. Dette gir For (x, y

Detaljer

Eksamensoppgave i TMA4320 Introduksjon til vitenskapelige beregninger

Eksamensoppgave i TMA4320 Introduksjon til vitenskapelige beregninger Institutt for matematiske fag Eksamensoppgave i TMA432 Introduksjon til vitenskapelige beregninger Faglig kontakt under eksamen: Anton Evgrafov Tlf: 453 163 Eksamensdato: 8. august 217 Eksamenstid (fra

Detaljer

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT UNIVERSITETET I OSLO ØKONOMISK INSTITUTT Bokmål Eksamen i: ECON1210 Forbruker, bedrift og marked Exam: ECON1210 Consumer Behaviour, Firm behaviour and Markets Eksamensdag: 12.12.2014 Sensur kunngjøres:

Detaljer

EKSAMENSOPPGAVE I SØK 1002 INNFØRING I MIKROØKONOMISK ANALYSE

EKSAMENSOPPGAVE I SØK 1002 INNFØRING I MIKROØKONOMISK ANALYSE Norges teknisk-naturvitenskapelige universitet Institutt for samfunnsøkonomi EKSAMENSOPPGAVE I SØK 1002 INNFØRING I MIKROØKONOMISK ANALYSE Faglig kontakt under eksamen: Hans Bonesrønning Tlf.: 9 17 64

Detaljer

PSi Apollo. Technical Presentation

PSi Apollo. Technical Presentation PSi Apollo Spreader Control & Mapping System Technical Presentation Part 1 System Architecture PSi Apollo System Architecture PSi Customer label On/Off switch Integral SD card reader/writer MENU key Typical

Detaljer

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT

UNIVERSITETET I OSLO ØKONOMISK INSTITUTT UNIVERSITETET I OSLO ØKONOMISK INSTITUTT Bokmål Eksamen i: ECON30 Økonomisk aktivitet og økonomisk politikk Exam: ECON30 Macroeconomic theory and policy Eksamensdag: 26.05. 204 Sensur kunngjøres: 6.06.204

Detaljer

UNIVERSITETET I OSLO

UNIVERSITETET I OSLO UNIVERSITETET I OSLO Det matematisk-naturvitenskapelige fakultet Eksamen i INF 3230 Formell modellering og analyse av kommuniserende systemer Eksamensdag: 4. april 2008 Tid for eksamen: 9.00 12.00 Oppgavesettet

Detaljer

Exploratory Analysis of a Large Collection of Time-Series Using Automatic Smoothing Techniques

Exploratory Analysis of a Large Collection of Time-Series Using Automatic Smoothing Techniques Exploratory Analysis of a Large Collection of Time-Series Using Automatic Smoothing Techniques Ravi Varadhan, Ganesh Subramaniam Johns Hopkins University AT&T Labs - Research 1 / 28 Introduction Goal:

Detaljer

GEOV219. Hvilket semester er du på? Hva er ditt kjønn? Er du...? Er du...? - Annet postbachelor phd

GEOV219. Hvilket semester er du på? Hva er ditt kjønn? Er du...? Er du...? - Annet postbachelor phd GEOV219 Hvilket semester er du på? Hva er ditt kjønn? Er du...? Er du...? - Annet postbachelor phd Mener du at de anbefalte forkunnskaper var nødvendig? Er det forkunnskaper du har savnet? Er det forkunnskaper

Detaljer