Tail Estimation of the Spectral Density under Fixed-Domain Asymptotics

Size: px
Start display at page:

Download "Tail Estimation of the Spectral Density under Fixed-Domain Asymptotics"

Transcription

1 Tail Estiation of the Spectral Density under Fixed-Doain Asyptotics Wei-Ying Wu, Chae Young Li and Yiin Xiao Wei-Ying Wu, Departent of Statistics & Probability Michigan State University, East Lansing, MI e-ail: Chae Young Li, Departent of Statistics & Probability Michigan State University, East Lansing, MI e-ail: Yiin Xiao, Departent of Statistics & Probability, Michigan State University, East Lansing, MI e-ail: Abstract: Consider a stationary Gaussian rando field on R d with the spectral density fλ that satisfies fλ c Hλ θ as λ for soe nonsingular atrix H. The paraeters c and θ control the tail behavior of the spectral density. c is related to a icroergodic paraeter and θ is related to a fractal index. For data observed on a grid, we propose estiators of c and θ by iniizing an objective function, which can be viewed as a weighted local Whittle likelihood and study their asyptotic properties under fixed-doain asyptotics. AMS 2000 subject classifications: Priary 62M30; secondary 62H11. Keywords and phrases: fixed-doain asyptotics, fractal index, fractal diension, Gaussian rando fields, infill asyptotics, icroergodic, Whittle likelihood. 1. Introduction With recent advances in technology, we are facing enorous aount of data sets. When data sets are observed on a regular grid, spectral analysis is popular Supported by MSU grant 08-IRGP-1532 Supported in part by NSF grant DMS

2 Wu, Li & Xiao/Tail Estiation of the Spectral Density 2 due to fast coputation using the Fast Fourier Transfor. For exaple, paraeters of the spectral density of a stationary lattice process can be estiated using a Whittle likelihood [Whittle 1954], which is ore efficient in ters of coputation copared to the axiu likelihood ethod on a spatial doain. Spatial data on a grid often can be regarded as a realization of a rando field on a lattice. That is, for a rando field, Zs on R d, data are observed at ϕj for J d {1,, j}, where ϕ is a grid length. When ϕ is fixed, asyptotic properties of paraeter estiates on a spectral doain have been studied by any authors [see, e.g., Whittle 1954, Guyon 1982, 1995, Boissy et al and Guo et al. 2009]. For exaple, Guyon 1982 studied asyptotic properties of estiators using a Whittle likelihood or its variants when a paraetric odel is assued for the spectral density of a stationary process on a lattice. Guo et al studied asyptotic properties of estiators of long-range dependence paraeters for anisotropic spatial linear process using a local Whittle likelihood ethod in which a paraetric for near zero frequency is only assued. This is an extension of Robinson 1995 for tie series. For spatial data, however, it is often natural to assue that the data are observed on a bounded doain of interest. More observations on the bounded doain iplies that the distance between observations, ϕ, is decreasing as the nuber of observations increases. This sapling schee requires a different asyptotic fraework, called fixed-doain asyptotics [Stein 1999] or infill asyptotics [Cressie 1993]. The classical asyptotic fraework when the sapling distance is fixed i.e. ϕ is fixed is called increasing-doain asyptotics to differentiate fro fixed-doain asyptotics. It has been shown that the asyptotic results under fixed-doain asyptotics can be different fro the results under increasing-doain asyptotics [see, e.g., Mardia and Marshall 1984, Ying 1991, 1993, Zhang 2004]. For exaple, Zhang 2004 showed not all paraeters in the Matérn covariance function of a stationary Gaussian rando field on R d are consistently estiable when d is saller than or equal to 3, while a reparaeterized quantity of variance and scale paraeters can be estiated consistently by the axiu likelihood ethod. On the other hand, the axiu likelihood estiators MLEs of variance and scale paraeters for a stationary Gaussian process under increasing-doain asyptotics are consistent and asyptotically noral [Mardia and Marshall 1984]. Although not all paraeters can be estiated consistently under fixeddoain asyptotics, a icroergodic paraeter can be estiated consistently [see, e.g., Ying 1991, 1993, Zhang 2004, Zhang and Zieran 2005, Du et al. 2009, Anderes 2010]. The icroergodicity of functions of paraeters

3 Wu, Li & Xiao/Tail Estiation of the Spectral Density 3 deterines the equivalence of probability easures and a icroergodic paraeter is the quantity that affects asyptotic ean squared prediction error under fixed-doain asyptotics. [Stein 1990a, 1990b, 1999]. Although there have been ore asyptotic results available recently under fixed-doain asyptotics, it is still a very few in contrast with vast literature on increasing-doain asyptotics. Also, ost results are for specific odels of covariance functions. For exaple, Ying 1991, 1993 and Chen et al studied asyptotic properties of estiators for a icroergodic paraeter in the exponential covariance function, while Zhang 2004, Loh 2005, Kaufan et al. 2008, Du et al and Anderes 2010 investigated asyptotic properties of estiators for the Matérn covariance function. Moreover, these asyptotic results are established in the spatial doain. Asyptotic work in the spectral doain are even less under fixed-doain asyptotics. Stein 1995 studied asyptotic properties of a spatial periodogra of a filtered version of a stationary Gaussian rando field. Li and Stein 2008 extended results of Stein 1995 and showed asyptotic norality of a soothed spatial cross-periodogra under fixed-doain asyptotics. Regarding the paraeter estiation in the spectral doain under fixed-doain asyptotics, Chan et al proposed a periodogra-based estiator of the fractal diension of a stationary Gaussian rando field when d = 1. In this paper, we propose estiators of paraeters that control the tail behavior of the spectral density for a stationary Gaussian rando field when the data are observed on a grid within a bounded doain and study their asyptotic properties under fixed-doain asyptotics. Let fλ be the spectral density of a stationary Gaussian rando field, Zs on R d and we assue that f λ c Hλ θ as λ, 1.1 where is a usual Euclidean nor, H is a nonsingular atrix and θ > d to ensure integrability of f. That is, we assue a power law for the tail behavior of the spectral density and do not assue any specific paraetric for of the spectral density. Also, this assuption allows a wide range of anisotropic spectral densities by introducing H. The proposed estiators are obtained by iniizing an objective function that can be viewed as a weighted local Whittle likelihood, in which Fourier frequencies near a pre-specified non-zero frequency are considered. This approach is siilar to the local Whittle likelihood ethod introduced by Robinson 1995 for estiating a long-range dependence paraeter in tie series analysis. For a stationary lattice process, Robinson 1995 proposed to estiate a long-range dependence paraeter by iniizing a Whittle

4 Wu, Li & Xiao/Tail Estiation of the Spectral Density 4 likelihood over Fourier frequencies near zero since the long-range dependence paraeter controls the behavior of the spectral density near zero. Meanwhile, we are interested in estiating paraeters that govern the spectral density of a rando field when the frequency is very large so that we need to focus on Fourier frequencies that are away fro zero. We establish consistency and asyptotic norality of an estiator of c and an estiator of θ, respectively, when the other paraeter is known. The paraeter c is related to a icroergodic paraeter. For exaple, consider the Matérn spectral density given as fλ = σ 2 α 2ν. 1.2 π d/2 α 2 + λ 2 ν+d/2 The Matérn spectral density has three paraeters, σ 2, α, ν, where σ 2 is the variance paraeter, α is the scale paraeter and ν is the soothness paraeter. Since the Matérn spectral density satisfies fλ σ2 α 2ν π d 2 λ 2ν+d as λ, we have c σ 2 α 2ν /π d/2 and θ 2ν+d, and σ 2 α 2ν is a icroergodic paraeter. Thus, estiating σ 2 α 2ν when ν is known is equivalent to estiate c when θ is known. There are several references that investigate estiation of σ 2 α 2ν in the spatial doain. Zhang 2004 showed that σ 2 and α can be estiated only in the for of σ 2 α 2ν under fixed-doain asyptotics when ν is known and d 3. Du et al investigated asyptotic properties of the MLE and a tapered MLE of σ 2 α 2ν when ν is known, α is fixed and d = 1 for a stationary Gaussian rando field. Anderes 2010 proposed an increent-based estiator of σ 2 α 2ν for a geoetric anisotropic Matérn covariance function and showed that α can be estiated separately when d > 4. The paraeter θ is related to the fractal index or fractal diension. For exaple, for a stationary Gaussian rando field, Zs, s R d, suppose that its covariance function Ct satisfies Ct C0 k t α as t for soe k and 0 < α 2. In this case, α is the fractal index that governs the roughness of saple paths of a rando field and the fractal diension D becoes D = d + 1 α/2. This follows fro Adler 1981, Chapter 8 or Theore 5.1 in Xue and Xiao 2010 where ore general results are proven for anisotropic Gaussian rando fields. When α = 2, it is possible that the saple

5 Wu, Li & Xiao/Tail Estiation of the Spectral Density 5 function ay be differentiable. This can be deterined by the soothness of Ct or in ters of the spectral easure of Zs see, e.g., Adler and Taylor 2007 or Xue and Xiao 2010 for further inforation. By an Abelian type theore, 1.3 holds if the corresponding spectral density satisfies fλ k λ α+d as λ so that θ α + d in our settings. There is a nuber of references that construct estiators based on fractal properties of processes. For exaple, Constantine and Hall 1994 estiated effective fractal diension using variogra for a non-gaussian stationary process on R. Chan and Wood 2004 introduced an increent-based estiator for the fractal diension of a stationary Gaussian rando field on R d with d = 1 or 2. Copared to these works in the spatial doain, the work by Chan et al to estiate the fractal diension is done in the spectral doain. In Section 2, we explain our settings and assuptions, and in Section 3 introduce our estiators and state the ain theores for the asyptotic properties of the proposed estiators. Section 4 discusses soe issues related to our approach and possible extension of the current work. All proofs are given in Appendices. 2. Preliinaries In this paper, we consider a stationary Gaussian rando field, Zs on R d with the spectral density fλ that satisfies 1.1. Define a lattice process Y ϕ J by Y ϕ J ZϕJ, where J Z d, the set of d-diensional integer-valued vectors. The corresponding spectral density of Y ϕ J is f ϕ λ = ϕ λ + 2πQ d f, ϕ Q Z d for λ π, π ] d. fϕ λ has a peak near the origin which is getting higher as ϕ 0. This causes a proble to estiate the spectral density using the periodogra [Stein 1995]. To alleviate the proble, we consider a discrete Laplacian operator to difference the data, which is proposed by Stein The Laplacian operator is defined by ϕ Zs = d {Zs + ϕ e j 2Zs + Zs ϕ e j }, where e j is the unit vector whose jth entry is 1. Depending on the behavior of the spectral density at high frequencies, we need to apply the Laplacian operator

6 Wu, Li & Xiao/Tail Estiation of the Spectral Density 6 iteratively to control the peak near the origin. Define Y τ ϕ J ϕ τ Zs as the lattice process obtained by applying the Laplacian operator τ ties. Then its corresponding spectral density becoes f ϕ τ λ = d 4 sin 2 λ j /2 The liit of f τ ϕ λ as ϕ 0 after scaling by ϕd θ is d ϕ d θ τ f ϕ λ c 4 sin 2 λ j /2 2τ 2τ f ϕ λ. 2.1 Q Z d Hλ + 2πQ θ for λ 0. Define for λ π, π] d, d g c,θ λ = c 4 sin 2 λj 2 2τ Q Z d Hλ + 2πQ θ I λ 0, 2.2 where I A = 1 if A is true and zero, otherwise. The liit function, g c,θ λ is integrable by choosing τ such that 4τ θ > d. When d = 1, siple differencing is preferred as discussed in Stein Then, 4τ will be replaced with 2τ in our results in Section 3. Now suppose that Zs is observed on the lattice ϕj. More specifically, we assue that we observe Yϕ τ J at J T = {1,..., } d after differencing Zs using the Laplacian operator τ ties. We further assue that ϕ = 1 so that the nuber of observations increases within a fixed observation doain. The spectral density of Yϕ τ J can be estiated by a periodogra which is defined using a discrete Fourier transfor of the data. That is, periodogra is defined by I τ λ = 2π d Dλ 2, where Dλ, the discrete Fourier transfor of the data, is given as Dλ = J T Y τ ϕ J exp{ i λ T J}. We consider the periodogra only at Fourier frequencies, 2π 1 J for J T { 1/2,, /2 } d, where x is the largest integer not greater

7 Wu, Li & Xiao/Tail Estiation of the Spectral Density 7 than x. A soothed periodogra at Fourier frequencies is defined by 2πJ Î τ = 2πJ + K W h KI τ, K T with weights W h K given by W h K = Λ h 2πK/ L T Λ h 2πL/, 2.3 where Λ h s = 1 s h Λ I { s h} h for a syetric continuous function Λ on R d that satisfies Λs 0 and Λ0 > 0 and I A is the indicator function of the set A. The nor is defined by s = ax{ s 1, s 2,..., s d }. For positive functions a and b, aλ bλ for λ A eans that there exist constants C 1 and C 2 such that 0 < C 1 aλ/bλ C 2 < for all possible λ A. For asyptotic results in this paper, we consider the following assuption on the spectral density f λ. Assuption 1. For a stationary Gaussian rando field Zs on R d, the spectral density fλ satisfies A f λ c Hλ θ as λ, for soe c > 0, θ > d and a nonsingular atrix H, B f λ is twice differentiable and there exists a positive constant C such that for λ > C, fλ 1 + λ θ, fλ λ j 1 + λ θ+1 and fλ λ j λ k 1 + λ θ+2 for j, k = 1,..., d. 3. Main Results Asyptotic properties of a spatial periodogra and a soothed spatial periodogra under fixed-doain asyptotics were investigated by Stein 1995 and Li and Stein They assue that the spectral density f is twice differentiable and satisfies 2.4 for all λ R d, which iplies that the spectral

8 Wu, Li & Xiao/Tail Estiation of the Spectral Density 8 density fλ behaves like 1 + λ θ for all λ. This is uch stronger condition than Assuption 1. This stronger condition allows to find asyptotic bounds for the expectation, variance and covariance of a spatial periodogra at Fourier frequency 2πJ/ by and J for J = 0. Consistency and asyptotic norality of a soothed spatial periodogra at Fourier frequency 2πJ/, however, are only available when li 2πJ/ = µ 0, that is, J should not be close to zero asyptotically. Since we ake use of asyptotic properties of a soothed spatial periodogra at such Fourier frequency, we extend those results in Stein 1995 and Li and Stein 2008 under Assuption 1. We focus on only a soothed spatial periodogra in the following theore, but results for a soothed spatial cross-periodogra can be shown siilarly. Throughout p d the paper, let denote the convergence in probability and denote the convergence in distribution. Theore 3.1. Suppose that the spectral density f of a stationary Gaussian rando field Zs on R d satisfies Assuption 1. Also suppose that 4τ > θ 1 and h = C γ for soe C > 0 where γ satisfies ax{d 2/d, 0} < γ < 1. Further, assue that li 2πJ/ = µ and 0 < µ < π. Let η = d1 γ/2. Then, we have and Î τ 2πJ/ f τ ϕ 2πJ/ η d θ Î τ 2πJ/ g c,θ µ p d N 0, Λ 2 Λ 2 1 d 2π g 2 C c,θµ, 3.2 where Λ r = [ 1,1] d Λ r sds. Reark 3.1. g c,θ is integrable under 4τ > θ d which is satisfied by the condition 4τ > θ 1. 4τ > θ 1 is necessary to show E Îτ 2πJ/ / f ϕ τ 2πJ/ 1 and the condition ax{d 2/d, 0} < γ < 1 is needed to show V ar Îτ 2πJ/ / f ϕ τ 2πJ/ 0 so that 3.1 can be shown. To estiate paraeters, c and θ, we consider the following objective function to be iniized. Lc, θ = { W h K log d θ g c,θ 2πJ + K/ 3.3 K T + 1 d θ I τ 2πJ + K/ g c,θ 2πJ + K/ },

9 Wu, Li & Xiao/Tail Estiation of the Spectral Density 9 where W h K is given in 2.3. In Lc, θ, 2πJ/ is any given Fourier frequency that satisfies J so that 2πJ/ is away fro 0. Lc, θ can be viewed as a weighted local Whittle likelihood. If Λ is a nonzero constant function, W h K 1/ K for K K, where K = {K T : 2πK/ h} and K is the nuber of eleents in the set K. Then, Lc, θ is the for of a local Whittle likelihood for the lattice data {Y τ δ J, J T } in which the true spectral density is replaced with d θ g c,θ. Note that g c,θ λ is the liit of the spectral density of Y τ δ J after scaling by d θ for non-zero λ when ϕ = 1. The suation in Lc, θ is over the Fourier frequencies near 2πJ/ by letting h 0 as. While a local Whittle likelihood ethod to estiate a long-range dependence paraeter for tie series considers Fourier frequencies near zero, we consider Fourier frequencies near a pre-specified nonzero frequency. For exaple, by choosing J such that 2πJ/ = π/2 1 d, where 1 d is the d-diensional vector of ones, Lc, θ considers frequencies only near π/21 d. For the estiation of c, we iniize Lc, θ with a known θ. Thus, the proposed estiator of c when θ is known as θ 0 is given by ĉ = arg in c C Lc, θ 0, where C is the paraeter space of c. ĉ has the explicit expression obtained by Lc, θ 0 / c = 0 : ĉ = K T W h K 1 d θ 0 I τ 2πJ + K/ g 0 2πJ + K/, 3.4 where g 0 g 1,θ0. The following theore establishes the consistency and asyptotic norality of the estiator ĉ. Theore 3.2. Suppose that the spectral density f of a stationary Gaussian rando field Zs on R d satisfies Assuption 1. Also suppose that 4τ > θ 0 1 for a known θ 0 and h = C γ for soe C > 0 where γ satisfies d/d + 2 < γ < 1. Further, assue that J satisfies 2πJ/ = π/2 1 d and the true paraeter c is in the interior of the paraeter space C which is a closed interval. Let η = d1 γ/2. Then, for ĉ given in 3.4, we have ĉ p c, 3.5 and η ĉ c d N 0, c 2 Λ 2 Λ 2 1 d 2π, 3.6 C where Λ r = [ 1,1] d Λ r sds.

10 Wu, Li & Xiao/Tail Estiation of the Spectral Density 10 Reark 3.2. We can prove Theore 3.2 for J such that li 2πJ/ = µ and 0 < µ < π instead of the specific choice of 2πJ/ = π/21 d, which we have chosen for siplicity of the proof. Reark 3.3. Without uch difficulty, Theore 3.2 can be also proved when we replace θ 0 with a consistent estiator ˆθ as long as the estiator ˆθ satisfies ˆθ θ 0 = o p log 1 p which iplies ˆθ θ 0 1. When we choose Λ as a constant function and C = 1/2π 2, we have η ĉ c d N 0, 2 d c 2 π d. For the Matérn spectral density given in 1.2 with d = 1, Du et al showed that for any fixed α 1 with known ν, the MLE of σ 2 satisfies n 1/2 ˆσ 2 α 2ν 1 σ 2 0α 2ν 0 d N 0, 2σ 2 0α 2ν 0 2, 3.7 where n is the saple size, and σ0 2 and α 0 are true paraeters. Note that is the saple size of Yϕ τ which is the τ ties differenced lattice process of Zs so that = n 2τ for the siple differencing and = n 4τ for the Laplace differencing. Since π 1/2 c = σ 2 α 2ν for d = 1, we have the sae asyptotic variance as in 3.7. However, our approach has a slower convergence rate since η < 1/3 when d = 1 as we used partial inforation. This is also the case for a local Whittle likelihood ethod in Robinson To estiate θ, we assue that c is known as c 0. The proposed estiator of θ is then given by ˆθ = arg in θ Θ Lc 0, θ, 3.8 where Θ is the paraeter space of θ. The consistency and the convergence rate of the proposed estiator ˆθ are given in the following Theore. Theore 3.3. Suppose that the spectral density f of a stationary Gaussian rando field Zs on R d satisfies Assuption 1. Also suppose that 4τ > θ 1 and h = C γ for soe C > 0 where γ satisfies d/d + 2 < γ < 1. Further, assue that J satisfies 2πJ/ = π/2 1 d and the true paraeter θ is in the interior of the paraeter space Θ which is a closed interval. Then, for ˆθ given in 3.8, we have ˆθ p θ. 3.9 In addition, ˆθ θ = o p log

11 Wu, Li & Xiao/Tail Estiation of the Spectral Density 11 Reark 3.4. The consistency of ˆθ is not enough to prove the asyptotic distribution of ˆθ since we have θ in the exponent of in the expression of Lc, θ. For the proof of the asyptotic distribution, we need the rate of convergence given in Fro Theore 3.3, we can now show the following Theore for the asyptotic distribution of ˆθ. Theore 3.4. Under the conditions of Theore 3.3, we have log η d ˆθ θ N 0, Λ d 2 2π Λ 2, 1 C where η = d1 γ/2. Reark 3.5. Note that we have a different convergence rate for ˆθ copared to the convergence rate for ĉ given in Theore 3.2. The additional ter log is fro the fact that θ is in the exponent of in the expression of Lc, θ. 4. Discussion We proposed estiators of c and θ that govern the tail behavior of the spectral density of a stationary Gaussian rando field on R d. The proposed estiators are obtained by iniizing the objective function given in 3.3. As entioned in Section 3, this objective function is siilar to the one used in the local Whittle likelihood ethod when a kernel function Λ in W h K is constant. When we replace d θ g c,θ with f τ ϕ λ and reove W hk in the expression given in 3.3, it can be thought of a Whittle approxiation to the likelihood of Y τ ϕj. This approxiation, however, has not been verified under fixed-doain asyptotics. One ight think that we can apply a siilar technique to prove the validity of a Whittle approxiation to the likelihood under increasing-doain asyptotics since Y τ ϕj is a lattice process. However, the spectral density f τ ϕ λ of Y τ ϕj converges to zero as ϕ 0, which requires a different approach and further investigation is needed. The weights in 3.3 is controlled by h, a bandwidth, which can be interpreted as a proportion of Fourier frequencies to be considered in the objective function. In our theores, we assue h = C γ for soe constant C. In proofs, we ake use of the properties of a soothed spatial periodogra Îτ. Thus, we could find the optial bandwidth that iniizes the ean squared error of Îτ. However, finding the ean squared error of Îτ needs explicit first order asyptotic expressions of the bias and variance of Îτ λ, which are not yet available. It will

12 Wu, Li & Xiao/Tail Estiation of the Spectral Density 12 be ore useful when we can estiate c and θ together or estiate θ when c is unknown. Due to the for of g c,θ, proving their asyptotic properties under fixed-doain asyptotics is challenging but worthwhile investigating further. The Assuption 1 A iplies that the spectral density is regularly varying at infinity with exponent θ. Together with a soothness condition Assuption 1 B, it satisfies assuptions A1 and A2 in Stein 2002, which includes slowly varying tail behavior. Assuptions A1 and A2 in Stein 2002 guarantee that there is a screening effect, that is, one can get a nearly optial predictor at a location s based on the observations nearest to s see Stein 2002 for further details. Then, our result that one can estiate tail behavior using only local inforation can be seen as a kind of analogue to a screening effect. Appendix A: The properties of g c,θ λ Soe properties of g c,θ λ are discussed in this Appendix. These properties will be used in the proofs given in Appendix B. Recall that 2τ d g c,θ λ = c 4 sin 2 λ j /2 Hλ + 2πQ θ Q Z d for λ 0. For a function g c,θ λ, let g denote the gradient of g with respect to λ and let ġ and g denote the first and second derivatives of g c,θ λ with respect to θ, respectively. That is, g = g/ λ 1,, g/ λ d, ġ = g c,θ λ/ θ and g = 2 g c,θ λ/ θ 2. LetA ρ = [ π, π] d \ ρ, ρ d for a fixed ρ that satisfies 0 < ρ < 1. Since we assue that the paraeter space Θ is a closed interval in Section 3, let Θ = [θ L, θ U ] and θ L > d. Although Lea A.1 can be shown for any fixed ρ with 0 < ρ < 1, we further assue that ρ is sall enough so that all Fourier frequencies near π/21 d considered in Lc, θ are contained in A ρ. Lea A.1. The following properties hold for g c,θ λ. a g c,θ λ is continuous on Θ A ρ. b There exist K L and K U such that for all θ, λ Θ A ρ, 0 < K L g c,θ λ K U <. A.1 c There exist K L and K U such that for all λ A ρ and all θ 1, θ 2 Θ, 0 < K L g c,θ1 λ/g c,θ2 λ K U <. A.2

13 Wu, Li & Xiao/Tail Estiation of the Spectral Density 13 d g, ġ, g, ġ/g and ġ/g are uniforly bounded on Θ A ρ. Proof. We prove the Lea when H is an identity atrix for siplicity. The results with a general nonsingular atrix H can be followed without uch difficulty. To show a, it is enough to show continuity of Q Z λ + d 2πQ θ on Θ A ρ { d 2τ since 4 sin2 λ j /2} is continuous on Aρ. It can be easily shown that Q Z d, Q >n λ + 2πQ θ converges to zero uniforly on Θ A ρ as n, which iplies the unifor convergence of Q Z d, Q n λ + 2πQ θ to g 1,θ λ. Thus, continuity of g c,θ λ follows fro the continuity of λ + 2πQ θ. To prove the reaining parts, we find the upper and lower bounds of Q Z d λ + 2πQ θ. For all θ, λ Θ A ρ, we have Q Z d λ + 2πQ θ π θu > 0 and Q Z d λ + 2πQ θ λ + 2πQ θ L + ϵ θ U Q Z d \{0} 2π d ϵ d θ L /d θ L + ϵ θ U, where the last inequality follows fro λ + 2πQ θ L λ + 2πy θ L dy Q Z d \0 y 1 2π d z θ L dz z ϵ = 2π d x d 1 x θ L dx = 2π d ϵ d θ L /θ L d, A.3 since θ L > d. Thus, we have 0 < k L λ + 2πQ θ k U <, A.4 Q Z d x ϵ where k L = π θ U and k U = 2π d ϵ d θ L /θ L d + ϵ θ U. Then, b follows fro A.4, 2τ d 4d sin 2 ϵ/2 2τ 4 sin 2 hλ j /2 4d 2τ,

14 Wu, Li & Xiao/Tail Estiation of the Spectral Density 14 and by setting K L c 4d sin 2 ϵ/2 2τ k L and K U c 4d 2τ k U. c follows fro observing that Q Zd λ + 2πQ θ has bounds that are unifor on Θ A ρ as given in A.4. lower and upper For d, we have g { d } 2τ 1 λ i = c 4τ 4 sin 2 λ j /2 sinλ i λ + 2πQ θ Q Z d { d } 2τ 1 θ 4 sin 2 λ j /2 λ i + 2πQ i λ + 2πQ θ 2 Q Z d K Q Z d λ + 2πQ θ K k U for soe constant K > 0 and k U given in A.4, which iplies unifor boundedness of g on Θ A ρ. For the unifor bound of ġ and g, we first copute ġ and g: ġ = { d } 2τ c 4 sin 2 λ j /2 λ + 2πQ θ log λ + 2πQ, Q Z d g = { d } 2τ c 4 sin 2 λ j /2 λ + 2πQ θ log λ + 2πQ 2. Q Z d Since we can find x 0 and K such that for a given β > 0, log x Kx β for all x > x 0, we can show that there exist n 0, K 1 and K 2 that satisfy ġ K 1 + K 2 Q Z d, Q n 0 λ + 2πQ θ+β for soe fixed β > 0. When we choose β = θ L θ/2, we can show that Q Z d, Q n 0 λ + 2πQ θ+β < using a siilar arguent to show A.3, which leads to unifor boundedness of ġ. Siilarly, we can show unifor boundedness of g. The unifor boundedness of ġ/g follows fro unifor boundedness of ġ and

15 Wu, Li & Xiao/Tail Estiation of the Spectral Density 15 b. To show unifor boundedness of ġ/g, consider Q Z ġ/g = λ + d 2πQ θ 2 λ i + 2πQ i 1 θ log λ + 2πQ λ i Q Z λ + 2πQ θ d Q Z λ + 2πQ θ log λ + 2πQ d + Q Z d λ + 2πQ θ 2 θ λ + 2πQ θ 2 λ i + 2πQ i Q Z d Since denoinators in the expression of ġ/g / λ i have unifor lower bounds as shown in A.4, it is enough to find unifor bounds of nuerators to show unifor boundedness of ġ/g / λ i. By observing that λ i +2πQ i λ + 2πQ and λ + 2πQ 1 K for soe K > 0 on A ρ, we can show that each nuerator in the expression of ġ/g / λ i is uniforly bounded on Θ A ρ using a siilar arguent to show unifor boundedness of ġ. Appendix B: Proofs of Theores in Section 3 Proof of Theore 3.1. If fλ satisfies 2.4 for all λ, 3.1 and 3.2 hold by results in Stein 1995 and Li and Stein To prove 3.1 and 3.2 when 2.4 holds only for large λ, we need to show that the effect of fλ on λ C is negligible. Consider a spectral density kλ which satisfies kλ c Hλ θ as λ and kλ is twice differentiable and satisfies 2.4 for all λ. Also assue that kλ fλ for λ > C. Let I f,τ λ be the periodogra at λ fro the observations under fλ and Note that a f,τ,ϕ J, K = 2π d d R d { d } 2τ 4 sin 2 ϕλ j /2 fλ sin 2 ϕλ j /2 sin ϕλ j /2 + πj j / sin ϕλ j /2 + πk j / dλ. E I f,τ 2πJ/ = a f,τ,ϕ J, J, V ar I f,τ 2πJ/ = a f,τ,ϕ J, J2 + a f,τ,ϕ J, J2.

16 Wu, Li & Xiao/Tail Estiation of the Spectral Density and 3.2 follow fro Theores 3, 6 and 12 in Li and Stein 2008 when these Theores hold for f under Assuption 1. The key part of proofs of these Theores under Assuption 1 is to show E I f,τ 2πJ/ f ϕ τ 2πJ/ = 1 + O β 1 B.1 V ar I f,τ 2πJ/ f τ ϕ 2πJ/2 = 1 + O β2, B.2 for soe β 1, β 2 > 0. Once B.1 and B.2 are shown, the other parts of proofs are siilar to the proofs in Li and Stein Since results in Stein 1995 and Li and Stein 2008 hold for kλ, we have B.1 and B.2 for kλ. Then, B.1 and B.2 for fλ follow fro a f,τ,ϕ J, ±J ak,τ,ϕ J, ±J = O d 4τ, B.3 for J that satisfies J and 2J/ Z d. B.3 holds since a f,τ,ϕ J, ±J ak,τ,ϕ J, ±J { d } 2τ = 2π d 4 sin 2 ϕλ j /2 fλ kλ λ C d sin 2 ϕλ j 2 sin ϕλ j /2 + πj j / sin ϕλ j /2 ± πj j / dλ 2π d d v d 4τ λ C { d } 2τ 4 sin 2 ϕλ j /2 fλ kλ sin 2 ϕλ j /2 sin ϕλ j /2 + πj j / sin ϕλ j /2 ± πj j / dλ for soe positive constant v since kλ fλ for λ > C and ϕλ j /2 ± πj j / stays away fro zero and π when is large. Proof of Theore 3.2. To show weak consistency of ĉ, we consider upper and lower bounds of ĉ. Let K U = argax K T,W h K 0 g 0 2πJ + K/ and K L = argin K T,,W h K 0 g 0 2πJ + K/. Recall that g 0 = g 1,θ0. Then,

17 Wu, Li & Xiao/Tail Estiation of the Spectral Density 17 we have K T W h KI2πJ τ + K/ d θ 0 g0 2πJ + K U ĉ / K T W h KI2πJ τ + K/ d θ 0 g0 2πJ + K L, / which can be rewritten as c Îτ 2πJ/ d θ0 g c,θ0 2πJ + K U / ĉ c Îτ 2πJ/ d θ 0 gc,θ0 2πJ + K L / B.4 with probability one. Note that both g c,θ0 2πJ + K U / and g c,θ0 2πJ + K L / converge to g c,θ0 π/21 d by continuity of g c,θ λ and d θ0 Î2πJ/ τ converges to g c,θ0 π/21 d in probability by Theore 3.1. Thus, it follows that ĉ converges to c in probability. For the asyptotic distribution of ĉ, note that we have Îτ η 2πJ/ g d θ0 c,θ0 π/21 d d N 0, Λ 2 Λ 2 1 d 2π gc,θ 2 C 0 π/21 d B.5 fro Theore 3.1 and η g c,θ0 2πJ + K E / g c,θ0 π/21 d 0, B.6 for E = U or L, since 4τ > θ 0 1, h = C γ and follows fro B.5 and B.6. d d+2 < γ < 1. Then, 3.6 To prove Theore 3.3, we consider following leas. Lea B.1. Consider a function h x = logx+d x 1, where d is a positive function of a positive integer. Also assue that d 1 as. Then, for a given r with 0 < r < 1, there exist δ r > 0 and M r > 0 such that for all M r, h x > δ r, for any x Z r {z : z 1 > r, z > 0}.

18 Wu, Li & Xiao/Tail Estiation of the Spectral Density 18 Proof. It can be easily shown that for any positive integer, h x is a convex function on 0, and iniized at x = 1/d with h 1/d 0. Let h x = logx + x 1. Since d 1, for any r 0, 1, there exists M r > 0 such that for all M r, we have 1/d 1 r and in{h 1 r, h 1+r} > 1/2 in{h 1 r, h 1 + r} > 0. Hence for all x Z r, we have h r x in{h 1 r, h 1 + r} > 1/2 in{h 1 r, h 1 + r} δ r. Lea B.2. For a positive integer and θ Θ, we have where A = log Lc 0, θ Lc 0, θ 0 A + B + C, θ θ 0 gc0,θ B = log 0 2πJ + S / 2πJ + S M / C = g 2πJ +S c0,θ 0 Î τ 2πJ + B.7 g 2πJ +S c0,θ d θ 0 gc0 2πJ +K M,θ 0 g 2πJ +S θ θ c0,θ 0 0 1, g 2πJ +S c0,θ Î τ 2πJ/ d θ0 2πJ + K M / In B.7-B.9, K M, K, S M and S are defined as, B.8 g c0,θ2πj + S M / g c0,θ2πj + S / 1 g c 0,θ 0 2πJ + K M / 2πJ + K / K M = arg ax {K T,W h K 0} 2πJ + K/, K = arg in {K T,W h K 0} 2πJ + K/, S M = arg ax {K T,W h K 0} log gc0,θ 0 2πJ + K/ g c0,θ2πj + K/ g c0,θ S = arg in 0 2πJ + K/ {K T,W h K 0} g c0,θ2πj + K/. Furtherore,, B.9. sup B = o1, B.10 θ Θ C = o p 1, B.11 where B.11 holds under the conditions of Theore 3.3.

19 Wu, Li & Xiao/Tail Estiation of the Spectral Density 19 Proof. Fro the expression of Lc, θ given in 3.3, we have Lc 0, θ Lc 0, θ 0 = { W h K log d θ g c0,θ 2πJ + K/ K T log d θ 0 2πJ + K/ } + 1 I W h K 2πJ τ + K/ d θ g c0 K T,θ2πJ + K/ 1 I W h K 2πJ τ + K/ d θ0 g c0,θ K T 0 2πJ + K/ = W h K log θ θ g 0 c 0,θ 0 2πJ + K/ g c0,θ 2πJ + K/ K T + I W h K 2πJ τ + K/ d θ0 g c0,θ K T 0 2πJ + K/ θ θ g 0 c 0,θ 0 2πJ + K/ g c0,θ2πj + K/ I W h K 2πJ τ + K/ d θ 0 gc0,θ K T 0 2πJ + K/ log θ θ g 0 c 0,θ 0 2πJ + S M / g c0,θ 2πJ + S M / + I W h K 2πJ τ + K/ d θ 0 gc0,θ K T 0 2πJ + K M / = log g θ θ0 c 0,θ 0 2πJ + S / g c0,θ2πj + S / I2πJ τ + K/ d θ0 2πJ + K / W h K K T g θ θ0 c 0,θ 0 2πJ + S M / g c0,θ2πj + S M / + Î2πJ/ τ d θ0 2πJ + K M / g c0,θ2πj + S / g c 0,θ 0 2πJ + K M / 2πJ + K / θ θ g 0 c 0,θ 0 2πJ + S / =: H.

20 Wu, Li & Xiao/Tail Estiation of the Spectral Density 20 H is further decoposed as H = log θ θ g 0 c 0,θ 0 2πJ + S / g c0,θ2πj + S / θ θ 0 g c 0,θ 0 2πJ + S / Î τ + 2πJ/ d θ 0 gc0,θ 0 2πJ + K M / gc0,θ + log 0 2πJ + S / g c0,θ2πj + S M / g c0,θ 0 2πJ + S M / g c0,θ2πj + S / + Î τ 2πJ/ d θ 0 gc0,θ 0 2πJ + K M / g c0,θ2πj + S / 1 1 g c 0,θ 0 2πJ + K M / 2πJ + K / which is A + B + C given in B.7-B.9. Note that 2πJ + K M /, 2πJ + K /, 2πJ + S M / and 2πJ + S / converge to π/21 d as. Note also that the convergence of 2πJ +S M / and 2πJ +S / holds for θ uniforly on Θ, because h 0. The continuity of g c0,θ in Lea A.1 iplies that gc0,θ log 0 2πJ + S / g c0,θ2πj + S M / 0 B.12 g c0,θ 0 2πJ + S M / g c0,θ2πj + S / holds for θ uniforly on Θ, therefore, sup Θ B = o1. Also, we have, d θ 0 Î τ 2πJ//g c0,θ 0 2πJ + K M / p 1, since d θ0 Î τ 2πJ//g c0,θ 0 π/21 d converges to one in probability by Theore 3.1 and 2πJ + K M / converges to π/21 d. Thus, together with 1 g c 0,θ 0 2πJ + K M / 2πJ + K / 0, B.13 C converges to zero in probability. Proof of Theore 3.3. Let Ω, F, P be the probability space where a stationary Gaussian rando field Zs is defined. To ephasize dependence on, we use ˆθ instead of ˆθ in this proof. Note that we have P Lc 0, ˆθ Lc 0, θ 0 0 = 1 B.14 for each positive integer by the definition of ˆθ. We are going to prove consistency of ˆθ by deriving a contradiction to B.14 when ˆθ does not converge

21 Wu, Li & Xiao/Tail Estiation of the Spectral Density 21 to θ 0 in probability. Suppose that ˆθ does not converge to θ 0 in probability. Then, there exist ϵ > 0, δ > 0 and M 1 such that for M 1, P ˆθ θ 0 > ϵ > δ. We define D = {ω Ω : ˆθ ω θ 0 > ϵ}. By Lea B.2, we have Lc 0, ˆθ ω Lc 0, θ 0 A + B + C, where A, B and C are given in B.7-B.9 with θ = ˆθ ω, ω D. We are going to show that there exist { k }, a subsequence of {} and a subset of D k on which A k + B k + C k is bounded away fro zero for large enough k. Note that A = h ˆθ θ 0 2πJ + S /, g c0,ˆθ 2πJ + S / where h is defined in Lea B.1 with d = Î τ 2πJ/ d θ 0 gc0,θ 0 2πJ + K M /, B.15 where K M is defined in Lea B.2. By Theore 3.1 and the convergence of g c0,θ 0 2πJ + K M / to g c0,θ 0 π/21 d, p we have d 1. Then, there exists {k }, a subsequence of {} such that d k converges to one alost surely. By B.13 in the proof of Lea B.2, alost sure convergence of d k iplies that C k defined in B.9 converges to zero alost surely. To use Lea B.1, we need unifor convergence of d k. By Egorov s Theore Folland, 1999, there exists G δ Ω such that d k and C k converge uniforly on G δ and P G δ > 1 δ/2. Let H k = D k G δ. Note that P H k > δ/2 > 0 for k M 1. On the other hand, there exists a M 2, which does not depend on ω, such that for k M 2, ˆθ k θ 0 2πJ + S k / k k g c0,ˆθ k 2πJ + S k / k 1 > 1 B.16 2 for all ω D k, because of the unifor boundedness of /g c0,θ. Then, by Lea B.1 with r = 1/2, there exist δ r > 0 and M r ax{m 1, M 2 } such that

22 Wu, Li & Xiao/Tail Estiation of the Spectral Density 22 for k M r, A k = log ˆθ k θ 0 k 2πJ + S k / k g c0,ˆθ k 2πJ + S k / k B.17 + > δ r d θ0 Î τ k 2πJ/ k k g c0,θ 0 2πJ + K M / k k θ 0 g c0,θ 0 2πJ + S k / k ˆθ k g c0,ˆθ k 2πJ + S k / k 1 uniforly on H k. By the unifor convergence of B on Θ shown in Lea B.2, there exists a M 3 such that for k M 3, B k < δ r 4 with θ = ˆθ k ω uniforly on H k. The unifor convergence of C k allows us to find M 4 such that for k M 4, B.18 on G δ C k < δ r 4 B.19 uniforly on H k. Therefore, for k ax{m r, M 3, M 4 }, we have A k + B k + C k A k B k C k > δ r /2 on H k which leads Lc 0, ˆθ k Lc 0, θ 0 > δ r 2 B.20 on H k. Since P H k > δ/2 > 0, it contradicts to B.14 which copletes the proof. Here, we do not need P k H k > 0 since B.14 should holds for any > 0. To show the convergence rate of ˆθ given in 3.10, it is enough to show that ˆθ θ p 0 1 which is equivalent to show that 2πJ + S / g c0,ˆθ 2πJ + S / ˆθ θ 0 g c0,θ 0 2πJ + S / g c0,ˆθ 2πJ + S / p 1, B.21 p 1. B.22 B.21 follows fro the consistency of ˆθ and the continuity of g c0,θ shown in Lea A.1. To show B.22, we consider a siilar arguent to show consistency

23 Wu, Li & Xiao/Tail Estiation of the Spectral Density 23 of ˆθ. For siplicity, we reset notations such as r, δ, δ r, M and D, etc. used in the proof of consistency. Suppose that B.22 does not hold. Then, there exists r > 0, δ > 0 and M 1 such that P ˆθ θ 0 2πJ + S / g c0,ˆθ 2πJ + S / 1 > r > δ for all M 1. On the other hand, there exists { k }, a subsequence of {}, such that d k 1, B k 0 and C k 0 alost surely, where d is given in B.15, B and C are given in B.8 and B.9 with θ = ˆθ ω. Then, by Egorov s Thoere, there exists Ω δ Ω such that P Ω δ > 1 δ/2 and d k, B k and C k are uniforly convergent on Ω δ. As in Lea B.1, for a k, a nonzero solution of h k b k = 0, where b = ˆθ θ 0 g c0,θ 0 2πJ + S / g c0,ˆθ 2πJ + S /, there exists M 2 such that a k 1 r uniforly on Ω δ for all k M 2. Now, define { } D = ω : ˆθ θ 0 2πJ + S / g c0,ˆθ 2πJ + S / 1 > r. Note that P D k Ω δ δ/2 > 0 for all k ax{m 1, M 2 }. Siilarly to the proof of Lea B.1, for each k ax{m 1, M 2 }, there exists δ r > 0 such that A k > δ r for all ω D k Ω δ. This iplies that P A k > δ r δ/2 for each k ax{m 1, M 2 }. Note that δ r does not depend on k which can be seen in Lea B.1. Meanwhile, there exists M 3 such that for k M 3, B k δ r /4, C k δ r /4 for all ω Ω δ. Hence we have P Lc 0, ˆθ Lc 0, θ 0 > δ r /2 δ/2 for k ax{m 1, M 2, M 3 }, which contradicts to B.14. Thus, B.22 is proved. Note that an alternative proof of the consistency of ˆθ is available [Wu 2011]. To proof Theore 3.4, we consider the following Lea.

24 Wu, Li & Xiao/Tail Estiation of the Spectral Density 24 Lea B.3. Under the conditions of Theore 3.3, let η = d1 γ/2, we have a b η I W h K 2πJ τ + K/ d θ 0 gc0,θ K T 0 2πJ + K/ 1 D N 0, Λ d 2 2π Λ 2, 1 C I τ 2πJ +K W h K 1 K T d θ0 g 2πJ +K c0,θ 0 ġ 2πJ +K c0,θ 0 = O g 2πJ p η +K c0,θ 0 B.23 B.24 Proof. To prove B.23, we find the asyptotic distribution of its lower and upper bounds. It can be easily shown that L η I W h K 2πJ τ + K/ d θ 0 gc0,θ K T 0 2πJ + K/ 1 U, where L = η Î2πJ/ τ d θ 0 gc0,θ 0 2πJ + K M / 1, B.25 U = η Î2πJ/ τ d θ0 g c0,θ 0 2πJ + K / 1 B.26 with K M and K as defined in Lea B.2. We rewrite L as L = η Î2πJ/ τ d θ 0 gc0,θ 0 π/21 d 1 π/21 d 2πJ + K M / g c0,θ + 0 π/21 d g c0,θ 0 2πJ + K M / 1. By Lea A.1 and γ > d/d + 2, we have π/21 d g c0,θ 0 2πJ + K M / η π/21 d 2πJ + K M / 1 1, 0.

25 Wu, Li & Xiao/Tail Estiation of the Spectral Density 25 Thus, by Theore 3.1, Siilarly, we can show L U d N d N 0, Λ 2 Λ 2 1 0, Λ 2 Λ 2 1 d 2π. C d 2π. C The convergence of lower and upper bounds to the sae distribution iplies B.23. To show B.24, we rewrite the LHS of B.24 as 1 K T W h K I τ 2πJ + K/ d θ 0 gc0,θ 0 2πJ + K/ ġc0,θ 0 2πJ + K/ g c0,θ 0 2πJ + K/ = W h Kġc0,θ02πJ + K/ g c0,θ K T 0 2πJ + K/ ġc 0,θ 0 π/21 d π/21 d I W h K 2πJ τ + K/ d θ0 g c0,θ K T 0 2πJ + K/ ġc 0,θ 0 2πJ + K/ 2πJ + K/ ġc 0,θ 0 π/21 d. π/21 d By Lea A.1 and γ > d/d + 2, we can show that η W h Kġc0,θ02πJ + K/ g c0,θ K T 0 2πJ + K/ ġc 0,θ 0 π/21 d π/21 d 0. Also, it can be easily shown that L I W h K 2πJ τ + K/ ġ c0,θ 0 2πJ + K/ d θ0 g c0,θ K T 0 2πJ + K/ 2πJ + K/ U, where with L = U = Î τ 2πJ/ d θ 0 gc0,θ 0 π/21 d Î τ 2πJ/ d θ 0 gc0,θ 0 π/21 d π/21 d ġ c0,θ 0 2πJ + P / gc 2, 0,θ 0 2πJ + P / π/21 d ġ c0,θ 0 2πJ + P M / gc 2, 0,θ 0 2πJ + P M / ġ c0,θ P M = arg ax 0 2πJ + K/ {K T,W h K 0} gc 2 0,θ 0 2πJ + K/, ġ c0,θ P = arg in 0 2πJ + K/ {K T,W h K 0} gc 2 0,θ 0 2πJ + K/.

26 Wu, Li & Xiao/Tail Estiation of the Spectral Density 26 By Lea A.1, γ > d/d + 2 and Theore 3.1, we can show that 2 d η L ġc 0,θ 0 π/21 d d ġc0,θ N 0, 0 π/21 d Λ 2 2π g c0,θ 0 π/21 d g c0,θ 0 π/21 d Λ 2, 1 C 2 d η U ġc 0,θ 0 π/21 d d ġc0,θ N 0, 0 π/21 d Λ 2 2π. π/21 d π/21 d C This copletes the proof of B.24. Λ 2 1 Proof of Theore 3.4. Let L = L/ θ and L = 2 L/ θ 2. To show the asyptotic distribution of ˆθ, we consider the Taylor expansion of Lc 0, ˆθ around θ 0, Lc 0, ˆθ = Lc 0, θ 0 + Lc 0, θˆθ θ 0, where θ lies on the line segent between ˆθ and θ 0. Since Lc 0, ˆθ = 0, we have log η ˆθ θ 0 = log η Lc0, θ 1 Lc0, θ 0. Thus, it is enough to show log 1 η Lc0, θ 0 log 2 Lc0, θ d N 0, Λ 2 Λ 2 1 d 2π, B.27 C p 1. B.28 Since Lc 0, θ 0 = log + W h Kġc0,θ02πJ + K/ g c0,θ K T 0 2πJ + K/ W h KI2πJ τ + K/ K T log d θ 0 2πJ + K/ + d θ0ġ c0,θ 0 2πJ + K/ d θ0 g c0,θ 0 2πJ + K/ 2 I = log W h K 2πJ τ + K/ d θ 0 gc0,θ K T 0 2πJ + K/ 1 + I τ W h K 1 2πJ + K/ ġc0,θ 0 2πJ + K/ d θ0 g c0,θ K T 0 2πJ + K/ 2πJ + K/, we see that B.27 follows fro Lea B.3.

27 Wu, Li & Xiao/Tail Estiation of the Spectral Density 27 Next we prove B.28. After soe siplification, we have Lc 0, θ = log 2 I W h K 2πJ τ + K/ d θg K T c0, θ2πj + K/ 2 log W h K Iτ 2πJ + K/ġ c0, θ2πj + K/ d θg 2 2πJ + K/ K T c 0, θ + 2 I2πJ τ + K/ġ 2 c W h K 0 2πJ + K/, θ d θg 3 2πJ + K/ K T c 0, θ + I W h K 1 2πJ τ + K/ gc0, θ2πj + K/ d θg K T c0, θ2πj + K/ g c0, θ2πj + K/ ġ 2 c W h K 0, θ 2πJ + K/ g 2 2πJ + K/ K T c 0, θ =: E 1 + E 2, where E 1 is the first ter with log 2 and E 2 is the last four ters in the expression of Lc 0, θ. First, we want to show that It can be easily shown that log 2 E 1 p 1. B.29 LB log 2 E 1 U, where with L = U M = Î2πJ/ τ θ θ 0 π/21 d d θ 0 gc0,θ 0 π/21 d g c0, θ2πj + P M /, Î2πJ/ τ θ θ 0 g c0,θ 0 π/21 d d θ 0 gc0,θ 0 π/21 d g c0, θ2πj + P / P M = arg ax {K T,W h K 0}g c0, θ2πj + K/, P = arg in {K T,W h K 0}g c0, θ2πj + K/. By Theore 3.1, 3.10 in Theore 3.3 and Lea A.1, we can show that both L and U converge to one in probability, which in turn iplies B.29. In a siilar way, we can show that log 1 E 2 = O p 1. Thus, together with B.29, we can show B.28, which copletes the proof.

28 Wu, Li & Xiao/Tail Estiation of the Spectral Density 28 References [1] Adler, R. J The geoetry of rando fields. John Wiley & Sons Ltd., New York. [2] Adler, R. J. and Taylor, J. E Rando fields and geoetry. Springer. [3] Anderes, E On the consistent separation of scale and variance for Gaussian rando fields. Ann. Statist [4] Boissy, Y., Bhattacharyya, B.B., Li, X. and Richardson, G.D Paraeter estiates for fractional autoregressive spatial processes. Ann. Statist [5] Chan, G., Hall, P. and Poskitt, D.S Periodogra-based estiators of fractal properties. Ann. Statist [6] Chan, G. and Wood, A. T. A Estiation of fractal diension for a class of non-gaussian stationary processes and fields. Ann. Statist [7] Chen, H.-S., Sipson, D.G. and Ying, Z Infill asyptotics for a stochastic process odel with easureent error. Statist. Sinica [8] Constantine, A. G. and Hall, P Characterizing surface soothness via estiation of effective fractal diension. J. R. Statist. Soc. B 56, [9] Cressie, N. A. C Statistics for spatial data rev. ed.. John Wiley, New York. [10] Du, J., Zhang, H. and Mandrekar, V. S Fixed-doain asyptotic properties of tapered axiu likelihood estiators. Ann. Statist [11] Folland, G. B Real analysis. Wiley-Interscience Publication, New York. [12] Guo, H., Li, C. and Meerschaert, M. M Local Whittle estiator for anisotropic rando fields. J. Multivariate Anal [13] Guyon, X Paraeter estiation for a stationary process on a d-diensional lattice. Bioetrika [14] Guyon, X Rando fields on a network : odeling, statistics, and applications. Springer-Verlag, New York. [15] Kaufan, C., Schervish, M. and Nychka, D Covariance tapering for likelihood-based estiation in large spatial datasets. J. Aer. Statist. Assoc [16] Loh, W.-L Fixed-doain asyptotics for a subclass of Matern-

29 Wu, Li & Xiao/Tail Estiation of the Spectral Density 29 type Gaussian rando fields. Ann. Statist [17] Li, C. and Stein, M. L Properties of spatial cross-periodogras using fixed-doain asyptotics. J. Multivariate Anal [18] Mardia, K. V. and Marshall, R. J Maxiu likelihood estiation of odels for residual covariance in spatial regression. Bioetrika [19] Robinson, P. M Gaussian seiparaetric estiation of long range dependence. Ann. Statist [20] Stein, M. L. 1990a. Unifor asyptotic optiality of linear predictions of a rando field using an incorrect second-order structure. Ann. Statist [21] Stein, M. L. 1990b. Bounds on the efficiency of linear predictions using an incorrect covariance function. Ann. Statist [22] Stein, M. L Fixed-doain asyptotics for spatial periodogras. J. Aer. Statist. Assoc [23] Stein, M. L Interpolation of spatial data. Springer, New York. [24] Stein, M. L The screening effect in kriging. Ann. Statist [25] Whittle, P On stationary processes in the plane. Bioetrika [26] Wu, W Tail estiation of the spectral density under fixed-doain asyptotics. Ph.D. Dissertation, Departent of Statistics and Probability, Michigan State University. [27] Xue, Y. and Xiao, Y Fractal and soothness properties of spacetie Gaussian odels. Frontiers Math. to appear. [28] Ying, Z Asyptotic properties of a axiu likelihood estiator with data fro a Gaussian process. J. Multivariate Anal [29] Ying, Z Maxiu likelihood estiation of paraeters under a spatial sapling schee. Ann. Statist [30] Zhang, H Inconsistent estiation and asyptotically equal interpolations in odel-based geostatistics. J. Aer. Statist. Assoc [31] Zhang, H. and Zieran, D. L Towards reconciling two asyptotic fraeworks in spatial statistics. Bioetrika 92,

TAIL ESTIMATION OF THE SPECTRAL DENSITY UNDER FIXED-DOMAIN ASYMPTOTICS. Wei-Ying Wu

TAIL ESTIMATION OF THE SPECTRAL DENSITY UNDER FIXED-DOMAIN ASYMPTOTICS. Wei-Ying Wu TAIL ESTIMATION OF THE SPECTRAL DENSITY UNDER FIXED-DOMAIN ASYMPTOTICS By Wei-Ying Wu A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of

More information

A Simple Regression Problem

A Simple Regression Problem A Siple Regression Proble R. M. Castro March 23, 2 In this brief note a siple regression proble will be introduced, illustrating clearly the bias-variance tradeoff. Let Y i f(x i ) + W i, i,..., n, where

More information

arxiv: v1 [math.pr] 17 May 2009

arxiv: v1 [math.pr] 17 May 2009 A strong law of large nubers for artingale arrays Yves F. Atchadé arxiv:0905.2761v1 [ath.pr] 17 May 2009 March 2009 Abstract: We prove a artingale triangular array generalization of the Chow-Birnbau- Marshall

More information

DEPARTMENT OF ECONOMETRICS AND BUSINESS STATISTICS

DEPARTMENT OF ECONOMETRICS AND BUSINESS STATISTICS ISSN 1440-771X AUSTRALIA DEPARTMENT OF ECONOMETRICS AND BUSINESS STATISTICS An Iproved Method for Bandwidth Selection When Estiating ROC Curves Peter G Hall and Rob J Hyndan Working Paper 11/00 An iproved

More information

Block designs and statistics

Block designs and statistics Bloc designs and statistics Notes for Math 447 May 3, 2011 The ain paraeters of a bloc design are nuber of varieties v, bloc size, nuber of blocs b. A design is built on a set of v eleents. Each eleent

More information

CS Lecture 13. More Maximum Likelihood

CS Lecture 13. More Maximum Likelihood CS 6347 Lecture 13 More Maxiu Likelihood Recap Last tie: Introduction to axiu likelihood estiation MLE for Bayesian networks Optial CPTs correspond to epirical counts Today: MLE for CRFs 2 Maxiu Likelihood

More information

13.2 Fully Polynomial Randomized Approximation Scheme for Permanent of Random 0-1 Matrices

13.2 Fully Polynomial Randomized Approximation Scheme for Permanent of Random 0-1 Matrices CS71 Randoness & Coputation Spring 018 Instructor: Alistair Sinclair Lecture 13: February 7 Disclaier: These notes have not been subjected to the usual scrutiny accorded to foral publications. They ay

More information

A Note on the Applied Use of MDL Approximations

A Note on the Applied Use of MDL Approximations A Note on the Applied Use of MDL Approxiations Daniel J. Navarro Departent of Psychology Ohio State University Abstract An applied proble is discussed in which two nested psychological odels of retention

More information

TEST OF HOMOGENEITY OF PARALLEL SAMPLES FROM LOGNORMAL POPULATIONS WITH UNEQUAL VARIANCES

TEST OF HOMOGENEITY OF PARALLEL SAMPLES FROM LOGNORMAL POPULATIONS WITH UNEQUAL VARIANCES TEST OF HOMOGENEITY OF PARALLEL SAMPLES FROM LOGNORMAL POPULATIONS WITH UNEQUAL VARIANCES S. E. Ahed, R. J. Tokins and A. I. Volodin Departent of Matheatics and Statistics University of Regina Regina,

More information

Learnability of Gaussians with flexible variances

Learnability of Gaussians with flexible variances Learnability of Gaussians with flexible variances Ding-Xuan Zhou City University of Hong Kong E-ail: azhou@cityu.edu.hk Supported in part by Research Grants Council of Hong Kong Start October 20, 2007

More information

Machine Learning Basics: Estimators, Bias and Variance

Machine Learning Basics: Estimators, Bias and Variance Machine Learning Basics: Estiators, Bias and Variance Sargur N. srihari@cedar.buffalo.edu This is part of lecture slides on Deep Learning: http://www.cedar.buffalo.edu/~srihari/cse676 1 Topics in Basics

More information

Non-Parametric Non-Line-of-Sight Identification 1

Non-Parametric Non-Line-of-Sight Identification 1 Non-Paraetric Non-Line-of-Sight Identification Sinan Gezici, Hisashi Kobayashi and H. Vincent Poor Departent of Electrical Engineering School of Engineering and Applied Science Princeton University, Princeton,

More information

Keywords: Estimator, Bias, Mean-squared error, normality, generalized Pareto distribution

Keywords: Estimator, Bias, Mean-squared error, normality, generalized Pareto distribution Testing approxiate norality of an estiator using the estiated MSE and bias with an application to the shape paraeter of the generalized Pareto distribution J. Martin van Zyl Abstract In this work the norality

More information

AN OPTIMAL SHRINKAGE FACTOR IN PREDICTION OF ORDERED RANDOM EFFECTS

AN OPTIMAL SHRINKAGE FACTOR IN PREDICTION OF ORDERED RANDOM EFFECTS Statistica Sinica 6 016, 1709-178 doi:http://dx.doi.org/10.5705/ss.0014.0034 AN OPTIMAL SHRINKAGE FACTOR IN PREDICTION OF ORDERED RANDOM EFFECTS Nilabja Guha 1, Anindya Roy, Yaakov Malinovsky and Gauri

More information

Support Vector Machine Classification of Uncertain and Imbalanced data using Robust Optimization

Support Vector Machine Classification of Uncertain and Imbalanced data using Robust Optimization Recent Researches in Coputer Science Support Vector Machine Classification of Uncertain and Ibalanced data using Robust Optiization RAGHAV PAT, THEODORE B. TRAFALIS, KASH BARKER School of Industrial Engineering

More information

Probability Distributions

Probability Distributions Probability Distributions In Chapter, we ephasized the central role played by probability theory in the solution of pattern recognition probles. We turn now to an exploration of soe particular exaples

More information

Extension of CSRSM for the Parametric Study of the Face Stability of Pressurized Tunnels

Extension of CSRSM for the Parametric Study of the Face Stability of Pressurized Tunnels Extension of CSRSM for the Paraetric Study of the Face Stability of Pressurized Tunnels Guilhe Mollon 1, Daniel Dias 2, and Abdul-Haid Soubra 3, M.ASCE 1 LGCIE, INSA Lyon, Université de Lyon, Doaine scientifique

More information

E0 370 Statistical Learning Theory Lecture 6 (Aug 30, 2011) Margin Analysis

E0 370 Statistical Learning Theory Lecture 6 (Aug 30, 2011) Margin Analysis E0 370 tatistical Learning Theory Lecture 6 (Aug 30, 20) Margin Analysis Lecturer: hivani Agarwal cribe: Narasihan R Introduction In the last few lectures we have seen how to obtain high confidence bounds

More information

Sharp Time Data Tradeoffs for Linear Inverse Problems

Sharp Time Data Tradeoffs for Linear Inverse Problems Sharp Tie Data Tradeoffs for Linear Inverse Probles Saet Oyak Benjain Recht Mahdi Soltanolkotabi January 016 Abstract In this paper we characterize sharp tie-data tradeoffs for optiization probles used

More information

Chaotic Coupled Map Lattices

Chaotic Coupled Map Lattices Chaotic Coupled Map Lattices Author: Dustin Keys Advisors: Dr. Robert Indik, Dr. Kevin Lin 1 Introduction When a syste of chaotic aps is coupled in a way that allows the to share inforation about each

More information

Course Notes for EE227C (Spring 2018): Convex Optimization and Approximation

Course Notes for EE227C (Spring 2018): Convex Optimization and Approximation Course Notes for EE227C (Spring 2018): Convex Optiization and Approxiation Instructor: Moritz Hardt Eail: hardt+ee227c@berkeley.edu Graduate Instructor: Max Sichowitz Eail: sichow+ee227c@berkeley.edu October

More information

Biostatistics Department Technical Report

Biostatistics Department Technical Report Biostatistics Departent Technical Report BST006-00 Estiation of Prevalence by Pool Screening With Equal Sized Pools and a egative Binoial Sapling Model Charles R. Katholi, Ph.D. Eeritus Professor Departent

More information

Model Fitting. CURM Background Material, Fall 2014 Dr. Doreen De Leon

Model Fitting. CURM Background Material, Fall 2014 Dr. Doreen De Leon Model Fitting CURM Background Material, Fall 014 Dr. Doreen De Leon 1 Introduction Given a set of data points, we often want to fit a selected odel or type to the data (e.g., we suspect an exponential

More information

ON THE TWO-LEVEL PRECONDITIONING IN LEAST SQUARES METHOD

ON THE TWO-LEVEL PRECONDITIONING IN LEAST SQUARES METHOD PROCEEDINGS OF THE YEREVAN STATE UNIVERSITY Physical and Matheatical Sciences 04,, p. 7 5 ON THE TWO-LEVEL PRECONDITIONING IN LEAST SQUARES METHOD M a t h e a t i c s Yu. A. HAKOPIAN, R. Z. HOVHANNISYAN

More information

The Distribution of the Covariance Matrix for a Subset of Elliptical Distributions with Extension to Two Kurtosis Parameters

The Distribution of the Covariance Matrix for a Subset of Elliptical Distributions with Extension to Two Kurtosis Parameters journal of ultivariate analysis 58, 96106 (1996) article no. 0041 The Distribution of the Covariance Matrix for a Subset of Elliptical Distributions with Extension to Two Kurtosis Paraeters H. S. Steyn

More information

Intelligent Systems: Reasoning and Recognition. Perceptrons and Support Vector Machines

Intelligent Systems: Reasoning and Recognition. Perceptrons and Support Vector Machines Intelligent Systes: Reasoning and Recognition Jaes L. Crowley osig 1 Winter Seester 2018 Lesson 6 27 February 2018 Outline Perceptrons and Support Vector achines Notation...2 Linear odels...3 Lines, Planes

More information

Pseudo-marginal Metropolis-Hastings: a simple explanation and (partial) review of theory

Pseudo-marginal Metropolis-Hastings: a simple explanation and (partial) review of theory Pseudo-arginal Metropolis-Hastings: a siple explanation and (partial) review of theory Chris Sherlock Motivation Iagine a stochastic process V which arises fro soe distribution with density p(v θ ). Iagine

More information

Fairness via priority scheduling

Fairness via priority scheduling Fairness via priority scheduling Veeraruna Kavitha, N Heachandra and Debayan Das IEOR, IIT Bobay, Mubai, 400076, India vavitha,nh,debayan}@iitbacin Abstract In the context of ulti-agent resource allocation

More information

On the block maxima method in extreme value theory

On the block maxima method in extreme value theory On the bloc axiethod in extree value theory Ana Ferreira ISA, Univ Tecn Lisboa, Tapada da Ajuda 349-7 Lisboa, Portugal CEAUL, FCUL, Bloco C6 - Piso 4 Capo Grande, 749-6 Lisboa, Portugal anafh@isa.utl.pt

More information

Estimating Parameters for a Gaussian pdf

Estimating Parameters for a Gaussian pdf Pattern Recognition and achine Learning Jaes L. Crowley ENSIAG 3 IS First Seester 00/0 Lesson 5 7 Noveber 00 Contents Estiating Paraeters for a Gaussian pdf Notation... The Pattern Recognition Proble...3

More information

Distributed Subgradient Methods for Multi-agent Optimization

Distributed Subgradient Methods for Multi-agent Optimization 1 Distributed Subgradient Methods for Multi-agent Optiization Angelia Nedić and Asuan Ozdaglar October 29, 2007 Abstract We study a distributed coputation odel for optiizing a su of convex objective functions

More information

The proofs of Theorem 1-3 are along the lines of Wied and Galeano (2013).

The proofs of Theorem 1-3 are along the lines of Wied and Galeano (2013). A Appendix: Proofs The proofs of Theore 1-3 are along the lines of Wied and Galeano (2013) Proof of Theore 1 Let D[d 1, d 2 ] be the space of càdlàg functions on the interval [d 1, d 2 ] equipped with

More information

Shannon Sampling II. Connections to Learning Theory

Shannon Sampling II. Connections to Learning Theory Shannon Sapling II Connections to Learning heory Steve Sale oyota echnological Institute at Chicago 147 East 60th Street, Chicago, IL 60637, USA E-ail: sale@athberkeleyedu Ding-Xuan Zhou Departent of Matheatics,

More information

Chapter 6 1-D Continuous Groups

Chapter 6 1-D Continuous Groups Chapter 6 1-D Continuous Groups Continuous groups consist of group eleents labelled by one or ore continuous variables, say a 1, a 2,, a r, where each variable has a well- defined range. This chapter explores:

More information

Support recovery in compressed sensing: An estimation theoretic approach

Support recovery in compressed sensing: An estimation theoretic approach Support recovery in copressed sensing: An estiation theoretic approach Ain Karbasi, Ali Horati, Soheil Mohajer, Martin Vetterli School of Coputer and Counication Sciences École Polytechnique Fédérale de

More information

Solutions of some selected problems of Homework 4

Solutions of some selected problems of Homework 4 Solutions of soe selected probles of Hoework 4 Sangchul Lee May 7, 2018 Proble 1 Let there be light A professor has two light bulbs in his garage. When both are burned out, they are replaced, and the next

More information

1 Bounding the Margin

1 Bounding the Margin COS 511: Theoretical Machine Learning Lecturer: Rob Schapire Lecture #12 Scribe: Jian Min Si March 14, 2013 1 Bounding the Margin We are continuing the proof of a bound on the generalization error of AdaBoost

More information

Supplementary Material for Fast and Provable Algorithms for Spectrally Sparse Signal Reconstruction via Low-Rank Hankel Matrix Completion

Supplementary Material for Fast and Provable Algorithms for Spectrally Sparse Signal Reconstruction via Low-Rank Hankel Matrix Completion Suppleentary Material for Fast and Provable Algoriths for Spectrally Sparse Signal Reconstruction via Low-Ran Hanel Matrix Copletion Jian-Feng Cai Tianing Wang Ke Wei March 1, 017 Abstract We establish

More information

Quantum algorithms (CO 781, Winter 2008) Prof. Andrew Childs, University of Waterloo LECTURE 15: Unstructured search and spatial search

Quantum algorithms (CO 781, Winter 2008) Prof. Andrew Childs, University of Waterloo LECTURE 15: Unstructured search and spatial search Quantu algoriths (CO 781, Winter 2008) Prof Andrew Childs, University of Waterloo LECTURE 15: Unstructured search and spatial search ow we begin to discuss applications of quantu walks to search algoriths

More information

Tail estimates for norms of sums of log-concave random vectors

Tail estimates for norms of sums of log-concave random vectors Tail estiates for nors of sus of log-concave rando vectors Rados law Adaczak Rafa l Lata la Alexander E. Litvak Alain Pajor Nicole Toczak-Jaegerann Abstract We establish new tail estiates for order statistics

More information

STRONG LAW OF LARGE NUMBERS FOR SCALAR-NORMED SUMS OF ELEMENTS OF REGRESSIVE SEQUENCES OF RANDOM VARIABLES

STRONG LAW OF LARGE NUMBERS FOR SCALAR-NORMED SUMS OF ELEMENTS OF REGRESSIVE SEQUENCES OF RANDOM VARIABLES Annales Univ Sci Budapest, Sect Cop 39 (2013) 365 379 STRONG LAW OF LARGE NUMBERS FOR SCALAR-NORMED SUMS OF ELEMENTS OF REGRESSIVE SEQUENCES OF RANDOM VARIABLES MK Runovska (Kiev, Ukraine) Dedicated to

More information

Nonlinear Log-Periodogram Regression for Perturbed Fractional Processes

Nonlinear Log-Periodogram Regression for Perturbed Fractional Processes Nonlinear Log-Periodogra Regression for Perturbed Fractional Processes Yixiao Sun Departent of Econoics Yale University Peter C. B. Phillips Cowles Foundation for Research in Econoics Yale University First

More information

Generalized eigenfunctions and a Borel Theorem on the Sierpinski Gasket.

Generalized eigenfunctions and a Borel Theorem on the Sierpinski Gasket. Generalized eigenfunctions and a Borel Theore on the Sierpinski Gasket. Kasso A. Okoudjou, Luke G. Rogers, and Robert S. Strichartz May 26, 2006 1 Introduction There is a well developed theory (see [5,

More information

Proc. of the IEEE/OES Seventh Working Conference on Current Measurement Technology UNCERTAINTIES IN SEASONDE CURRENT VELOCITIES

Proc. of the IEEE/OES Seventh Working Conference on Current Measurement Technology UNCERTAINTIES IN SEASONDE CURRENT VELOCITIES Proc. of the IEEE/OES Seventh Working Conference on Current Measureent Technology UNCERTAINTIES IN SEASONDE CURRENT VELOCITIES Belinda Lipa Codar Ocean Sensors 15 La Sandra Way, Portola Valley, CA 98 blipa@pogo.co

More information

arxiv: v1 [cs.ds] 3 Feb 2014

arxiv: v1 [cs.ds] 3 Feb 2014 arxiv:40.043v [cs.ds] 3 Feb 04 A Bound on the Expected Optiality of Rando Feasible Solutions to Cobinatorial Optiization Probles Evan A. Sultani The Johns Hopins University APL evan@sultani.co http://www.sultani.co/

More information

Fixed-to-Variable Length Distribution Matching

Fixed-to-Variable Length Distribution Matching Fixed-to-Variable Length Distribution Matching Rana Ali Ajad and Georg Böcherer Institute for Counications Engineering Technische Universität München, Gerany Eail: raa2463@gail.co,georg.boecherer@tu.de

More information

Asymptotic mean values of Gaussian polytopes

Asymptotic mean values of Gaussian polytopes Asyptotic ean values of Gaussian polytopes Daniel Hug, Götz Olaf Munsonius and Matthias Reitzner Abstract We consider geoetric functionals of the convex hull of norally distributed rando points in Euclidean

More information

Lecture 21. Interior Point Methods Setup and Algorithm

Lecture 21. Interior Point Methods Setup and Algorithm Lecture 21 Interior Point Methods In 1984, Kararkar introduced a new weakly polynoial tie algorith for solving LPs [Kar84a], [Kar84b]. His algorith was theoretically faster than the ellipsoid ethod and

More information

Lower Bounds for Quantized Matrix Completion

Lower Bounds for Quantized Matrix Completion Lower Bounds for Quantized Matrix Copletion Mary Wootters and Yaniv Plan Departent of Matheatics University of Michigan Ann Arbor, MI Eail: wootters, yplan}@uich.edu Mark A. Davenport School of Elec. &

More information

Polygonal Designs: Existence and Construction

Polygonal Designs: Existence and Construction Polygonal Designs: Existence and Construction John Hegean Departent of Matheatics, Stanford University, Stanford, CA 9405 Jeff Langford Departent of Matheatics, Drake University, Des Moines, IA 5011 G

More information

Statistics and Probability Letters

Statistics and Probability Letters Statistics and Probability Letters 79 2009 223 233 Contents lists available at ScienceDirect Statistics and Probability Letters journal hoepage: www.elsevier.co/locate/stapro A CLT for a one-diensional

More information

A Theoretical Analysis of a Warm Start Technique

A Theoretical Analysis of a Warm Start Technique A Theoretical Analysis of a War Start Technique Martin A. Zinkevich Yahoo! Labs 701 First Avenue Sunnyvale, CA Abstract Batch gradient descent looks at every data point for every step, which is wasteful

More information

Bootstrapping Dependent Data

Bootstrapping Dependent Data Bootstrapping Dependent Data One of the key issues confronting bootstrap resapling approxiations is how to deal with dependent data. Consider a sequence fx t g n t= of dependent rando variables. Clearly

More information

3.8 Three Types of Convergence

3.8 Three Types of Convergence 3.8 Three Types of Convergence 3.8 Three Types of Convergence 93 Suppose that we are given a sequence functions {f k } k N on a set X and another function f on X. What does it ean for f k to converge to

More information

AN EFFICIENT CLASS OF CHAIN ESTIMATORS OF POPULATION VARIANCE UNDER SUB-SAMPLING SCHEME

AN EFFICIENT CLASS OF CHAIN ESTIMATORS OF POPULATION VARIANCE UNDER SUB-SAMPLING SCHEME J. Japan Statist. Soc. Vol. 35 No. 005 73 86 AN EFFICIENT CLASS OF CHAIN ESTIMATORS OF POPULATION VARIANCE UNDER SUB-SAMPLING SCHEME H. S. Jhajj*, M. K. Shara* and Lovleen Kuar Grover** For estiating the

More information

Estimation of the Mean of the Exponential Distribution Using Maximum Ranked Set Sampling with Unequal Samples

Estimation of the Mean of the Exponential Distribution Using Maximum Ranked Set Sampling with Unequal Samples Open Journal of Statistics, 4, 4, 64-649 Published Online Septeber 4 in SciRes http//wwwscirporg/ournal/os http//ddoiorg/436/os4486 Estiation of the Mean of the Eponential Distribution Using Maiu Ranked

More information

Supplement to: Subsampling Methods for Persistent Homology

Supplement to: Subsampling Methods for Persistent Homology Suppleent to: Subsapling Methods for Persistent Hoology A. Technical results In this section, we present soe technical results that will be used to prove the ain theores. First, we expand the notation

More information

The Weierstrass Approximation Theorem

The Weierstrass Approximation Theorem 36 The Weierstrass Approxiation Theore Recall that the fundaental idea underlying the construction of the real nubers is approxiation by the sipler rational nubers. Firstly, nubers are often deterined

More information

M ath. Res. Lett. 15 (2008), no. 2, c International Press 2008 SUM-PRODUCT ESTIMATES VIA DIRECTED EXPANDERS. Van H. Vu. 1.

M ath. Res. Lett. 15 (2008), no. 2, c International Press 2008 SUM-PRODUCT ESTIMATES VIA DIRECTED EXPANDERS. Van H. Vu. 1. M ath. Res. Lett. 15 (2008), no. 2, 375 388 c International Press 2008 SUM-PRODUCT ESTIMATES VIA DIRECTED EXPANDERS Van H. Vu Abstract. Let F q be a finite field of order q and P be a polynoial in F q[x

More information

ESTIMATING AND FORMING CONFIDENCE INTERVALS FOR EXTREMA OF RANDOM POLYNOMIALS. A Thesis. Presented to. The Faculty of the Department of Mathematics

ESTIMATING AND FORMING CONFIDENCE INTERVALS FOR EXTREMA OF RANDOM POLYNOMIALS. A Thesis. Presented to. The Faculty of the Department of Mathematics ESTIMATING AND FORMING CONFIDENCE INTERVALS FOR EXTREMA OF RANDOM POLYNOMIALS A Thesis Presented to The Faculty of the Departent of Matheatics San Jose State University In Partial Fulfillent of the Requireents

More information

PRÜFER SUBSTITUTIONS ON A COUPLED SYSTEM INVOLVING THE p-laplacian

PRÜFER SUBSTITUTIONS ON A COUPLED SYSTEM INVOLVING THE p-laplacian Electronic Journal of Differential Equations, Vol. 23 (23), No. 23, pp. 9. ISSN: 72-669. URL: http://ejde.ath.txstate.edu or http://ejde.ath.unt.edu ftp ejde.ath.txstate.edu PRÜFER SUBSTITUTIONS ON A COUPLED

More information

Generalized AOR Method for Solving System of Linear Equations. Davod Khojasteh Salkuyeh. Department of Mathematics, University of Mohaghegh Ardabili,

Generalized AOR Method for Solving System of Linear Equations. Davod Khojasteh Salkuyeh. Department of Mathematics, University of Mohaghegh Ardabili, Australian Journal of Basic and Applied Sciences, 5(3): 35-358, 20 ISSN 99-878 Generalized AOR Method for Solving Syste of Linear Equations Davod Khojasteh Salkuyeh Departent of Matheatics, University

More information

STOPPING SIMULATED PATHS EARLY

STOPPING SIMULATED PATHS EARLY Proceedings of the 2 Winter Siulation Conference B.A.Peters,J.S.Sith,D.J.Medeiros,andM.W.Rohrer,eds. STOPPING SIMULATED PATHS EARLY Paul Glasseran Graduate School of Business Colubia University New Yor,

More information

Course Notes for EE227C (Spring 2018): Convex Optimization and Approximation

Course Notes for EE227C (Spring 2018): Convex Optimization and Approximation Course Notes for EE7C (Spring 018: Convex Optiization and Approxiation Instructor: Moritz Hardt Eail: hardt+ee7c@berkeley.edu Graduate Instructor: Max Sichowitz Eail: sichow+ee7c@berkeley.edu October 15,

More information

On the Use of A Priori Information for Sparse Signal Approximations

On the Use of A Priori Information for Sparse Signal Approximations ITS TECHNICAL REPORT NO. 3/4 On the Use of A Priori Inforation for Sparse Signal Approxiations Oscar Divorra Escoda, Lorenzo Granai and Pierre Vandergheynst Signal Processing Institute ITS) Ecole Polytechnique

More information

ANALYSIS OF A NUMERICAL SOLVER FOR RADIATIVE TRANSPORT EQUATION

ANALYSIS OF A NUMERICAL SOLVER FOR RADIATIVE TRANSPORT EQUATION ANALYSIS OF A NUMERICAL SOLVER FOR RADIATIVE TRANSPORT EQUATION HAO GAO AND HONGKAI ZHAO Abstract. We analyze a nuerical algorith for solving radiative transport equation with vacuu or reflection boundary

More information

Computational and Statistical Learning Theory

Computational and Statistical Learning Theory Coputational and Statistical Learning Theory Proble sets 5 and 6 Due: Noveber th Please send your solutions to learning-subissions@ttic.edu Notations/Definitions Recall the definition of saple based Radeacher

More information

Testing Properties of Collections of Distributions

Testing Properties of Collections of Distributions Testing Properties of Collections of Distributions Reut Levi Dana Ron Ronitt Rubinfeld April 9, 0 Abstract We propose a fraework for studying property testing of collections of distributions, where the

More information

Random Process Review

Random Process Review Rando Process Review Consider a rando process t, and take k saples. For siplicity, we will set k. However it should ean any nuber of saples. t () t x t, t, t We have a rando vector t, t, t. If we find

More information

Consistent Multiclass Algorithms for Complex Performance Measures. Supplementary Material

Consistent Multiclass Algorithms for Complex Performance Measures. Supplementary Material Consistent Multiclass Algoriths for Coplex Perforance Measures Suppleentary Material Notations. Let λ be the base easure over n given by the unifor rando variable (say U over n. Hence, for all easurable

More information

This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and

This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and This article appeared in a ournal published by Elsevier. The attached copy is furnished to the author for internal non-coercial research and education use, including for instruction at the authors institution

More information

This model assumes that the probability of a gap has size i is proportional to 1/i. i.e., i log m e. j=1. E[gap size] = i P r(i) = N f t.

This model assumes that the probability of a gap has size i is proportional to 1/i. i.e., i log m e. j=1. E[gap size] = i P r(i) = N f t. CS 493: Algoriths for Massive Data Sets Feb 2, 2002 Local Models, Bloo Filter Scribe: Qin Lv Local Models In global odels, every inverted file entry is copressed with the sae odel. This work wells when

More information

Learnability and Stability in the General Learning Setting

Learnability and Stability in the General Learning Setting Learnability and Stability in the General Learning Setting Shai Shalev-Shwartz TTI-Chicago shai@tti-c.org Ohad Shair The Hebrew University ohadsh@cs.huji.ac.il Nathan Srebro TTI-Chicago nati@uchicago.edu

More information

Constrained Consensus and Optimization in Multi-Agent Networks arxiv: v2 [math.oc] 17 Dec 2008

Constrained Consensus and Optimization in Multi-Agent Networks arxiv: v2 [math.oc] 17 Dec 2008 LIDS Report 2779 1 Constrained Consensus and Optiization in Multi-Agent Networks arxiv:0802.3922v2 [ath.oc] 17 Dec 2008 Angelia Nedić, Asuan Ozdaglar, and Pablo A. Parrilo February 15, 2013 Abstract We

More information

A Bernstein-Markov Theorem for Normed Spaces

A Bernstein-Markov Theorem for Normed Spaces A Bernstein-Markov Theore for Nored Spaces Lawrence A. Harris Departent of Matheatics, University of Kentucky Lexington, Kentucky 40506-0027 Abstract Let X and Y be real nored linear spaces and let φ :

More information

Support Vector Machines. Maximizing the Margin

Support Vector Machines. Maximizing the Margin Support Vector Machines Support vector achines (SVMs) learn a hypothesis: h(x) = b + Σ i= y i α i k(x, x i ) (x, y ),..., (x, y ) are the training exs., y i {, } b is the bias weight. α,..., α are the

More information

arxiv: v1 [stat.ot] 7 Jul 2010

arxiv: v1 [stat.ot] 7 Jul 2010 Hotelling s test for highly correlated data P. Bubeliny e-ail: bubeliny@karlin.ff.cuni.cz Charles University, Faculty of Matheatics and Physics, KPMS, Sokolovska 83, Prague, Czech Republic, 8675. arxiv:007.094v

More information

The degree of a typical vertex in generalized random intersection graph models

The degree of a typical vertex in generalized random intersection graph models Discrete Matheatics 306 006 15 165 www.elsevier.co/locate/disc The degree of a typical vertex in generalized rando intersection graph odels Jerzy Jaworski a, Michał Karoński a, Dudley Stark b a Departent

More information

Some Proofs: This section provides proofs of some theoretical results in section 3.

Some Proofs: This section provides proofs of some theoretical results in section 3. Testing Jups via False Discovery Rate Control Yu-Min Yen. Institute of Econoics, Acadeia Sinica, Taipei, Taiwan. E-ail: YMYEN@econ.sinica.edu.tw. SUPPLEMENTARY MATERIALS Suppleentary Materials contain

More information

e-companion ONLY AVAILABLE IN ELECTRONIC FORM

e-companion ONLY AVAILABLE IN ELECTRONIC FORM OPERATIONS RESEARCH doi 10.1287/opre.1070.0427ec pp. ec1 ec5 e-copanion ONLY AVAILABLE IN ELECTRONIC FORM infors 07 INFORMS Electronic Copanion A Learning Approach for Interactive Marketing to a Custoer

More information

New Bounds for Learning Intervals with Implications for Semi-Supervised Learning

New Bounds for Learning Intervals with Implications for Semi-Supervised Learning JMLR: Workshop and Conference Proceedings vol (1) 1 15 New Bounds for Learning Intervals with Iplications for Sei-Supervised Learning David P. Helbold dph@soe.ucsc.edu Departent of Coputer Science, University

More information

arxiv: v2 [math.st] 11 Dec 2018

arxiv: v2 [math.st] 11 Dec 2018 esting for high-diensional network paraeters in auto-regressive odels arxiv:803659v [aths] Dec 08 Lili Zheng and Garvesh Raskutti Abstract High-diensional auto-regressive odels provide a natural way to

More information

Optimal Jamming Over Additive Noise: Vector Source-Channel Case

Optimal Jamming Over Additive Noise: Vector Source-Channel Case Fifty-first Annual Allerton Conference Allerton House, UIUC, Illinois, USA October 2-3, 2013 Optial Jaing Over Additive Noise: Vector Source-Channel Case Erah Akyol and Kenneth Rose Abstract This paper

More information

ANALYSIS OF HALL-EFFECT THRUSTERS AND ION ENGINES FOR EARTH-TO-MOON TRANSFER

ANALYSIS OF HALL-EFFECT THRUSTERS AND ION ENGINES FOR EARTH-TO-MOON TRANSFER IEPC 003-0034 ANALYSIS OF HALL-EFFECT THRUSTERS AND ION ENGINES FOR EARTH-TO-MOON TRANSFER A. Bober, M. Guelan Asher Space Research Institute, Technion-Israel Institute of Technology, 3000 Haifa, Israel

More information

Detection and Estimation Theory

Detection and Estimation Theory ESE 54 Detection and Estiation Theory Joseph A. O Sullivan Sauel C. Sachs Professor Electronic Systes and Signals Research Laboratory Electrical and Systes Engineering Washington University 11 Urbauer

More information

Ch 12: Variations on Backpropagation

Ch 12: Variations on Backpropagation Ch 2: Variations on Backpropagation The basic backpropagation algorith is too slow for ost practical applications. It ay take days or weeks of coputer tie. We deonstrate why the backpropagation algorith

More information

On Constant Power Water-filling

On Constant Power Water-filling On Constant Power Water-filling Wei Yu and John M. Cioffi Electrical Engineering Departent Stanford University, Stanford, CA94305, U.S.A. eails: {weiyu,cioffi}@stanford.edu Abstract This paper derives

More information

Testing equality of variances for multiple univariate normal populations

Testing equality of variances for multiple univariate normal populations University of Wollongong Research Online Centre for Statistical & Survey Methodology Working Paper Series Faculty of Engineering and Inforation Sciences 0 esting equality of variances for ultiple univariate

More information

Optimal nonlinear Bayesian experimental design: an application to amplitude versus offset experiments

Optimal nonlinear Bayesian experimental design: an application to amplitude versus offset experiments Geophys. J. Int. (23) 155, 411 421 Optial nonlinear Bayesian experiental design: an application to aplitude versus offset experients Jojanneke van den Berg, 1, Andrew Curtis 2,3 and Jeannot Trapert 1 1

More information

In this chapter, we consider several graph-theoretic and probabilistic models

In this chapter, we consider several graph-theoretic and probabilistic models THREE ONE GRAPH-THEORETIC AND STATISTICAL MODELS 3.1 INTRODUCTION In this chapter, we consider several graph-theoretic and probabilistic odels for a social network, which we do under different assuptions

More information

Recovering Data from Underdetermined Quadratic Measurements (CS 229a Project: Final Writeup)

Recovering Data from Underdetermined Quadratic Measurements (CS 229a Project: Final Writeup) Recovering Data fro Underdeterined Quadratic Measureents (CS 229a Project: Final Writeup) Mahdi Soltanolkotabi Deceber 16, 2011 1 Introduction Data that arises fro engineering applications often contains

More information

Feature Extraction Techniques

Feature Extraction Techniques Feature Extraction Techniques Unsupervised Learning II Feature Extraction Unsupervised ethods can also be used to find features which can be useful for categorization. There are unsupervised ethods that

More information

Stochastic Subgradient Methods

Stochastic Subgradient Methods Stochastic Subgradient Methods Lingjie Weng Yutian Chen Bren School of Inforation and Coputer Science University of California, Irvine {wengl, yutianc}@ics.uci.edu Abstract Stochastic subgradient ethods

More information

Local asymptotic powers of nonparametric and semiparametric tests for fractional integration

Local asymptotic powers of nonparametric and semiparametric tests for fractional integration Stochastic Processes and their Applications 117 (2007) 251 261 www.elsevier.co/locate/spa Local asyptotic powers of nonparaetric and seiparaetric tests for fractional integration Xiaofeng Shao, Wei Biao

More information

COS 424: Interacting with Data. Written Exercises

COS 424: Interacting with Data. Written Exercises COS 424: Interacting with Data Hoework #4 Spring 2007 Regression Due: Wednesday, April 18 Written Exercises See the course website for iportant inforation about collaboration and late policies, as well

More information

Uniform Approximation and Bernstein Polynomials with Coefficients in the Unit Interval

Uniform Approximation and Bernstein Polynomials with Coefficients in the Unit Interval Unifor Approxiation and Bernstein Polynoials with Coefficients in the Unit Interval Weiang Qian and Marc D. Riedel Electrical and Coputer Engineering, University of Minnesota 200 Union St. S.E. Minneapolis,

More information

PAC-Bayes Analysis Of Maximum Entropy Learning

PAC-Bayes Analysis Of Maximum Entropy Learning PAC-Bayes Analysis Of Maxiu Entropy Learning John Shawe-Taylor and David R. Hardoon Centre for Coputational Statistics and Machine Learning Departent of Coputer Science University College London, UK, WC1E

More information

Kinetic Theory of Gases: Elementary Ideas

Kinetic Theory of Gases: Elementary Ideas Kinetic Theory of Gases: Eleentary Ideas 17th February 2010 1 Kinetic Theory: A Discussion Based on a Siplified iew of the Motion of Gases 1.1 Pressure: Consul Engel and Reid Ch. 33.1) for a discussion

More information

Fundamental Limits of Database Alignment

Fundamental Limits of Database Alignment Fundaental Liits of Database Alignent Daniel Cullina Dept of Electrical Engineering Princeton University dcullina@princetonedu Prateek Mittal Dept of Electrical Engineering Princeton University pittal@princetonedu

More information

Experimental Design For Model Discrimination And Precise Parameter Estimation In WDS Analysis

Experimental Design For Model Discrimination And Precise Parameter Estimation In WDS Analysis City University of New York (CUNY) CUNY Acadeic Works International Conference on Hydroinforatics 8-1-2014 Experiental Design For Model Discriination And Precise Paraeter Estiation In WDS Analysis Giovanna

More information