Browsing by Author "Lin, KY"

Now showing 1 - 6 of 6

A novel nonsymmetric K_-Lanczos algorithm for the generalized nonsymmetric K_-eigenvalue problems
(1997-02-01)
Ferng, WR
;
Lin, KY
;
Lin, WW
;
應用數學系
;
Department of Applied Mathematics
In this article, we present a novel algorithm, named nonsymmetric K---Lanczos algorithm, for computing a few extreme eigenvalues of the generalized eigenvalue problem Mx = lambda Lx, where the matrices M and L have the so-called K-+/--structures. We demonstrate a K---tridiagonalization procedure preserves the K-+/--structures. An error bound for the extreme K---Ritz value obtained from this new algorithm is presented. When compared with the class nonsymmetric Lanczos approach, this method has the same order of computational complexity and can be viewed as a special 2 x 2-block nonsymmetric Lanczos algorithm. Numerical experiments with randomly generated K---matrices show that our algorithm converges faster and more accurate than the nonsymmetric Lanczos algorithm. (C) Elsevier Science Inc., 1997
A novel nonsymmetric K_-Lanczos algorithm for the generalized nonsymmetric K_-eigenvalue problems
(1997-02-01)
Ferng, WR
;
Lin, KY
;
Lin, WW
;
應用數學系
;
Department of Applied Mathematics
In this article, we present a novel algorithm, named nonsymmetric K---Lanczos algorithm, for computing a few extreme eigenvalues of the generalized eigenvalue problem Mx = lambda Lx, where the matrices M and L have the so-called K-+/--structures. We demonstrate a K---tridiagonalization procedure preserves the K-+/--structures. An error bound for the extreme K---Ritz value obtained from this new algorithm is presented. When compared with the class nonsymmetric Lanczos approach, this method has the same order of computational complexity and can be viewed as a special 2 x 2-block nonsymmetric Lanczos algorithm. Numerical experiments with randomly generated K---matrices show that our algorithm converges faster and more accurate than the nonsymmetric Lanczos algorithm. (C) Elsevier Science Inc., 1997
Low-temperature series expansions for square-lattice Ising model with first and second neighbour interactions
(2002-12-20)
Kao, YM
;
Chen, M
;
Lin, KY
;
物理研究所
;
Institute of Physics
We have calculated the low-temperature series expansions of the spontaneous magnetization and the zero-field susceptibility of the square-lattice ferromagnetic Ising model with first-neighbour interaction J(1) and second-neighbour interaction J(2) to the 30th and 26th order respectively by computer. Our results extend the previous calculations by Lee and Lin to six more orders. We use the Pade approximants to estimate the critical exponents and the critical temperature for different ratios of R = J(2)/J(1). The estimated critical temperature as a function of R agrees with the estimation by Oitmaa from high-temperature series expansions.
New load-based resource allocation algorithms for packet scheduling in CDMA uplink
(2004)
Sheen, WH
;
Fu, IK
;
Lin, KY
;
電信工程研究所
;
Institute of Communications Engineering
Packet scheduling plays a key role in radio resource management of a wireless access network to best utilize the available radio resource. A radio-link packet scheduling consists of two steps: prioritization and resource allocation. Prioritization is a procedure that prioritizes users' order of service so as to guarantee QoS (quality of services), user's fairness and achieve as high as possible system throughput. Resource allocation, on the other hand, executes the "real" resource allocation under the constraints of physical limitations imposed by the radio segment of the access system. This paper proposes two new load-based resource allocation methods for packet scheduling in CDMA (code division multiple access) uplink with multiple services. One is called maximum-load based power allocation (MLBPA), and the other real-load based power allocation (RLBPA). MLBPA is easier to implement, but RLBPA has the better performance. By taking into consideration of the realistic channel conditions such as imperfect power control and transmit power variation, both of our methods outperform the current methods in terms of system throughput, frame loss rate and transmission delay. In addition, by monitoring the system load at all times, the planned coverage of the system can always be guaranteed.
Sampling time effect on 2,4-Toluene diisocyanate concentration by using different samplers
(2006)
Tsai, CJ
;
Lin, HC
;
Lin, KY
;
Shih, TS
;
Chang, KC
In order to understand the sampling time effect on the 2,4-TDI (Toluene diisocyanate) concentrations, laboratory and field tests were conducted in this study. An ADS (annular denuder sampler) and two OFFCs (open face filter holders) with different filters were tested for 1 to 120 minutes in the laboratory using 2,4-TDI gas. In the field study, the standard sampling method, the dual filter the triple filter and annular denuder systems were used at two workplaces to study the change of 2,4-TDI concentrations with sampling time from 15 to 60 minutes. The test results in both laboratory and field studies show that the sampling time influences the sampled TDI concentration considerably which may be due to reaction of TDI with water vapor and polyo in the sampling process. It is evident that as sampling time increases the TDI concentration decreases very significantly.
Spectral analysis of some iterations in the Chandrasekhar's H-functions
(2003-08-01)
Juang, J
;
Lin, KY
;
Lin, WW
;
應用數學系
;
Department of Applied Mathematics
Two very general, fast and simple iterative methods were proposed by Bosma and de Rooij (Bosma, P. B., de Rooij, W. A. (1983). Efficient methods to calculate Chandrasekhar's H functions. Astron. Astrophys. 126:283--292.) to determine Chandrasekhar's H-functions. The methods are based on the use of the equation 'j h = (F) over tilde (h), where (F) over tilde = ((f1) over tilde, (f2) over tilde, . . . (f(n)) over tilde")(T) is a nonlinear map from R-n to R-n. Here (f(i)) over tilde = 1 /(root1 - c + Sigma(k=1)(n) (c(k)mu(k)h(k)/mu(i) + mu(k)), 0 < c less than or equal to 1, i = 1,2, . . . ,n. One such method is essentially a nonlinear Gauss-Seidel iteration with respect to (F) over tilde. The other ingenious approach is to normalize each iterate after a nonlinear Gauss-Jacobi iteration with respect to (F) over tilde is taken. The purpose of this article is two-fold. First, we prove that both methods converge locally. Moreover, the convergence rate of the second iterative method is shown to be strictly less than (root3 - 1)/2. Second, we show that both the Gauss-Jacobi method and Gauss-Seidel method with respect to some other known alternative forms of the Chandrasekhar's H-functions either do not converge or essentially stall for c = 1.