On the Equivalence Between a Commonly Used Correlation Coefficient and a Least Squares Function
|
|
- Alvin Davis
- 7 years ago
- Views:
Transcription
1 On the Eqivalence Between a Commonly Used Correlation Coefficient and a Least Sqares Fnction Diane C. Jamrog George N. Phillips, Jr. Yin Zhang Janary 2003 (revised October 2003) Abstract Many objective fnctions have been proposed in X-ray crystallography to solve the moleclar replacement (MR) problem and other optimization problems. In this paper, we establish the eqivalence between optimizing two target fnctions: a commonly sed correlation coefficient and a least sqares fnction. This eqivalence may be in neighborhoods abot the global optima or the entire MR variable space depending on whether the average vales of the observed and calclated data are sbtracted from observed and calclated data. In addition, we also present an argment that the correlation coefficient between strctre factor magnitdes is likely to perform better than the correlation coefficient between intensities. This was confirmed by the MR program SOMoRe, especially when low-resoltion data were sed. 1 Introdction A major goal in X-ray crystallography is to qantitatively compare the observed and calclated diffraction patterns for a moleclar strctre being solved. This comparison Department of Comptational and Applied Mathematics, Rice University, Hoston, Texas, USA. Department of Biochemistry and Department of Compter Sciences, University of Wisconsin-Madison, Madison, Wisconsin, USA. 1
2 is sefl in the evalation of trial protein models, refinements of strctres, and error estimation. The measre of closeness between observed and calclated intensities (or strctre factor magnitdes) is determined by a target fnction. The choice of target fnction has been debated, and mch effort has been pt into developing new target fnctions. In particlar, solving the moleclar replacement (MR) problem is often a critical step in determining a detailed moleclar strctre. The MR problem is an optimization problem to determine the orientation and position of a model protein that prodce calclated intensities closest to those observed from a crystal with nknown atomic strctre. Varios target fnctions for the MR problem have been discssed. See [1, 2, 4, 6, 8, 11, 12, 14], for example. In this article, we establish that maximzing a correlation coefficient, which is a commonly sed objective fnction for the MR problem, is eqivalent to minimizing a least sqares fnction when the calclated intensities are properly scaled and some mild assmptions are met. In other words, the two respective optimization problems have the same set of global optimizers. 2 Objective fnctions We introdce a correlation coefficient and a least sqares fnction, and then we prove that the set of global optimizers of the respective optimization problems are identical nder some mild assmptions. 2.1 Correlation coefficient The Pearson correlation coefficient is often sed to solve the MR problem both becase it can be interpreted in terms of Patterson fnctions [3, 6] and becase it is scale invariant. In the X-ray crystallography literatre, this correlation coefficient is typically written as: C(I c (), I o h ( Ih c ) = () Ic () )( Ih o Io ) [ h ( Ih c() Ic () ) 2 ] 1/2 [ h ( Ih o Io ) 2 ] 1/2, (1) where I o h and I c h() are the observed and calclated intensities occrring at the lattice point h, R n specifies the orientation and translation of the model being positioned 2
3 by MR, h is the smmation over all h (or intensities) in the resoltion range, and I o and I c are the average vales of the observed and calclated intensities, respectively. Of corse, strctre factor magnitdes ( Fh c and Fh ) o can be sed in place of intensities; Fh c 2 = Ih c and F h o 2 = Ih o. The correlation coefficient can be written in more general terms as C(w(), w o ) = w() T w o w o w() = cos w(), wo, (2) where cos w(), w o is the cosine of the angle between the two vectors w() R m and w o R m. The vectors w() and w o are typically defined to be F c () k F c () k and F o k F o k, respectively. If k = 1, then strctre factor magnitdes are sed, and if k = 2, then intensities are sed. Ths, the correlation coefficient is scale invariant, becase the correlation coefficient is eqal to the cosine of an angle, and scaling either vector does not change the cosine of the angle between the two vectors. Obviosly, C(w(), w o ) [ 1, 1]. However, if the average vales are not sbtracted from the observed and calclated intensities, then C(w(), w o ) [0, 1] becase both w() and w o will be non-negative. Finally, the MR problem can be posed as 2.2 Least Sqares fnction min (1 C(w(), w o )). (3) A natral target fnction to measre the disagreement between the observed and calclated intensities is the following least sqares fnction: L(w(), α) = α w() w o 2, (4) where α R is a scale factor, w() R m is the vector of calclated intensities, and w o R m is the vector of observed intensities. In general, w() and w o can be either F c () k and F o k, respectively, or F c () k F c () k and F o k F o k, respectively, for k = 1 or 2. If the least sqares fnction is sed, then either the calclated or the observed intensities shold be scaled becase the observed intensities are measred on a relative scale dring the X-ray crystallography experiment. We choose to scale the calclated intensities, bt the the same effect can be achieved by scaling the observed 3
4 intensities by 1/α. As a reslt, the MR problem can be posed as the minimization of the disagreement between observed and calclated intensities over all possible linear scale factors and all possible orientations and translations of the model protein: min L(w(), α). (5), α The least sqares fnction is generally not sed as an objective fnction for the MR problem bt has been sed as an objective fnction for rigid body refinement, a comptational process that is often sed to refine a MR soltion. Most likely, the MR problem is not posed as (5) becase at face vale (5) may appear to be a more difficlt optimization problem than (3) de to the higher dimension of the variable space. However, the first two lemmas of the next section sggest the appropriate scale factor α. If this scale factor is sed, then optimizing the above least sqares fnction does not involve any more variables than optimizing the correlation coefficient. 3 Proof of Eqivalence In this section, we present for lemmas and a theorem establishing the eqivalence between minimizing L(w(), α) and 1 C(w(), w o ). In other words, (, α ) is a global minimizer of L(w(), α) if and only if is also a global minimizer of 1 C(w(), w o ). Two optimization problems will be referred to as eqivalent if the two sets of global optimizers are identical. This eqivalence will be symbolically denoted as. The seqence of theoretical reslts begins with a lemma that ses the reslt that if w() 0, then the optimal scale factor for the least sqares fnction is β() = w() T w o /(w() T w()). The second lemma establishes that minimizing L(w(), α) is eqivalent to minimizing L(w(), β()) when the above optimal scale factor β() is sed. The third lemma shows that 1 C 2 (w(), w o ) = L(w(), β())/ w o 2 provided w o 0 and w() 0. Finally, the theorem ties all these reslts together to show the eqivalence between minimizing L(w(), α) and 1 C(w(), w o ) nder mild assmptions. The role the assmptions play with respect to the regions of eqivalence are discssed following the theorem. 4
5 3.1 Theoretical reslts Lemma 1 For, v R m and 0, min α α R v 2 = T v 2 T v = v ( 2 1 cos 2, v ). (6) Proof: For fixed and v, the optimal scale factor is α = T v/( T ) or the soltion to the normal eqations for the minimization problem above. Now, sing this optimal scale factor, ( T v ) T ( T v ) T v T v = ( T ) 2 v T 2 T v T T T v + v T v, = v T v (T v) 2 T, ( = v T v 1 (T v) 2 ). v T v T Finally, sing the definition, cos, v = T v/( v ), v T v ( 1 (T v) 2 ) v T v T = v 2 ( 1 cos 2, v ). (7) Lemma 2 Let L(w(), α) be the least sqares fnction as defined in (4), where w o R m and w : R n R m and α R. Assme there exists R n sch that w() T w o > 0. (8) Then where min,α L(w(), α) min L(w(), β()), (9) β() = w()t w o w() 2. (10) Proof: Let f(, α) = α w() w o 2 and g(v) = β(v)w(v) w o 2. (11) 5
6 To prove the lemma, we show (, α ) U = { (ũ, α) sch that f(ũ, α) f(, α) (, α) R n R } (12) if and only if and V = { ṽ sch that g(ṽ) g(v) v R n } (13) α = β( ). (14) Let (, α ) U. Assmption (8) implies w( ) 0. Hence, as shown in Lemma 1, the niqe soltion to min γ γ w( ) w o 2 (15) is well defined as γ = w( ) T w o / w( ) 2 = β( ). Therefore, g( ) = β( ) w( ) w o 2 α w( ) w o 2 = f(, α ) β(v) w(v) w o 2, (16) that is, g( ) g(v) for arbitrary v. Ths, V. Moreover, f(, α ) = α w( ) w o 2 β( ) w( ) w o 2 = g( ), (17) becase (, α ) is a global minimizer of f(, α). Ths, α = β( ), since α w( ) w o 2 = β( ) w( ) w o 2 and β( ) is the niqe minimizer of (15). In addition, g( ) = f(, α ). Now, let v V, and sppose f(v, β(v )) > f(, α ). This ineqality implies g(v ) > g( ), a contradiction. Therefore, (v, β(v )) U. Lemma 3 Let C(w(), w o ) be the correlation fnction as defined in (2), where w o R m and w : R n R m. Let L(w(), α) be the least sqares fnction as defined in (4), and β() be the scale factor as defined in (10). Assme w() 0 and w o 0. Then 1 C 2 (w(), w o ) = L(w(), β()) w o 2. (18) Proof: Since w() 0, by Lemma 1, L(w(), β()) = w o 2 ( 1 cos 2 w(), w o ) = w o 2 ( 1 C 2 (w(), w o ) ). (19) 6
7 Therefore, becase w o 0, 1 C 2 (w(), w o ) = L(w(), β()) w o 2. (20) Lemma 4 Let C(w(), w o ) be the correlation fnction as defined in (2), where w o R m and w : R n R m is continos fnction on a compact set D R n. Assme γ 1 = min cos w(), w o, γ 2 = max cos w(), w o, and γ 1 < γ 2, (21) where the minimm and maximm are taken over the set D. Then over D Proof: Clearly, Similarly, min min 1 C 2 (w(), w o ) min 1 C(w(), w o ). (22) 1 C 2 (w(), w o ) max min 1 C(w(), w o ) max C 2 (w(), w o ) max cos 2 w(), w o. (23) cos w(), w o. (24) Now, given assmption (21), is a global maximizer of cos w(), w o if and only if cos w( ), w o = γ 2. Similarly, is a global maximizer of cos 2 w(), w o if and only if cos w( ), w o = γ 2 since γ 2 2 > γ2 1. Theorem 1 Let C(w(), w o ) be the correlation fnction as defined in (2), where w o R m and w : R n R m is a continos fnction on a compact set D R n. Let L(w(), α) be the least sqares fnction as defined in (4), and β() be the scale factor as defined in (10). Assme there exists R n sch that and w() T w o > 0, (25) γ 1 = min cos w(), w o, γ 2 = max cos w(), w o, and γ 1 < γ 2, (26) where the minimm and maximm are taken over the set D. Then over the set D min,α L(w(), α) min 1 C(w(), w o ). (27) 7
8 Proof: Given (25), w( ) 0, and by Lemma 2, Given (25), by Lemma 3, min,α L(w(), α) min L(w(), β()). (28) min L(w(), β()) min 1 C 2 (w(), w o ). (29) Finally, given (26), by Lemma 4, min 1 C 2 (w(), w o ) min 1 C(w(), w o ). (30) 3.2 Regions of eqivalence The assmptions of the lemmas and theorem are satisfied for the observed and calclated intensities, either in neighborhoods abot a global minimizer or for all in the MR variable space. First for the MR problem, assmption (25) shold always be satisfied becase w() = I c () 0 and w o = I o 0. The calclated and observed intensities shold not all be eqal to zero. Similarly, if w() = I c () I c () and w o = I o I o, then w() 0 becase the calclated intensities become less bright at a fairly rapid rate as their distance from the origin in reciprocal grows [13, p. 165]. For the same reason, w o 0. Second, whether assmption (26) holds for any in the optimization variable space D depends on the definition of w() and w o. (For example, in MR may be eqal to (θ 1, θ 2, θ 3, x, y, z) and D = [0, 2π] 3 [0, 1] 3.) Assmption (26) implies that γ 1 cos w(), w o γ 2, (31) where γ 1 < γ 2. If w() = I c () I c and w o = I o I o, then (26) may be satisfied only in a neighborhood of a global minimizer rather than for all. If the average vales are sbtracted, then the cosine of the angle between the two vectors, w() and w o may be large and violate assmption (26). However, if the model protein is accrate enogh, then in a neighborhood of the global minimizer, the initial 8
9 angle between the observed and calclated data shold be small enogh so that sbtracting the average vales will not increase the angle so mch as to violate (26). We now give a concrete example that shows that if the means are sbtracted, then there may be regions for which the fnction 1 C(w(), w o ) and the least sqares fnction are not eqivalent. Sppose the means are sbtracted and C(w(), w o ) has a local minimm at sch that C(w( ), w o ) < 0. Then, 1 C(w(), w o ) will have a local maximm at, bt w o 2 (1 C 2 (w(), w o )) = L(w(), β()) will have a local minimm at. Ths, optimization of the two fnctions will not be eqivalent near. In contrast, if the means are not sbtracted, then the cosine will always be non-negative, and assmption (26) and Lemma 4 will hold for all ; that is, eqivalence between the two fnctions will hold for the entire optimization variable space D. (Of corse, the above argments are the same if strctre factor magnitdes are sed in place of intensities.) Finally, we note that for the least sqares fnction there does not seem to be a jstification for sbtracting I o and I c from the observed and calclated intensities. On the other hand, when the correlation coefficient is sed as a traditional rotation fnction, Brnger notes that the nmerator of the correlation coefficient is eqal to the real space rotation fnction given some assmptions [3], and for the real space rotation fnction, the very large sprios origin peak is damped by sbtracting the average vales of the intensities; see [10, 13], for example. 4 Intensities verses strctre factor magnitdes Dring or development of the MR program SOMoRe [9], we sed sets of low-resoltion intensities and strctre factor magnitdes to compte srrogate fnctions, that is, fnctions that cold be sampled relatively qickly to identify regions of the MR variable space were soltions might exist. Based on or nmerical experimentation with SOMoRe, we feel that C( F o, F c ) is likely to be more accrate that C(I o, I c ) especially when lowresoltion data is sed. Similarly, Glykos and Kokkindis also advocate the general se of strctre factor magnitdes over intensities [7]. For example, dring some of SOMoRe s deterministic searches of the srrogate fnction, good starting points for local optimization cold not be fond when C(I o, I c ) was sed bt cold be fond when C( F o, F c ) was sed. (By good starting points, we mean starting 9
10 points that were sfficiently close to a global minimizer sch that the local optimization method BFGS cold converge to the soltions of the MR problems.) Besides the nmerical evidence presented in [9], we believe that C( F o, F c ) is more accrate becase this fnction essentially encorporates a weighting of the intensities according to the error in their measrements. The observation of the diffraction intensities dring an X-ray crystallography experiment is a stochastic process with nderlying Poisson statistics. Ths, the error in the measrements is proportional to the sqare root of the intensities. A proper point-wise weighting scheme of w h I h wold have a mltiplier of w h = 1/ I h, and this weighting effectively prodces C( F o, F c ) from C(I o, I c ). 10
11 Acknowledgments Diane Jamrog was spported in part by a training fellowship from the Keck Center for Comptational Biology (NSF GRT Grant BIR , NSF RTG BIR , and NLM 5 T15 LM07093) and NSF Grant DMS George N. Phillips, Jr. was spported by the National Institte of Health GM Yin Zhang was spported in part by DOE Grant DE-FG03-97ER25331, DOE/LANL Contract and NSF Grant DMS
12 References [1] T. Blndell and L. Johnson. Protein Crystallography. Academic Press, [2] J. Borge, C. Alvarex-Ra, and S. Garcia-Granda. A new vector-search rotation fnction: image seeking fnctions revisited in macromoleclar crystallography. Jornal of Moleclar Biology, D56: , [3] A. T. Brnger. Patterson correlation searches and refinement. Methods in Enzymology, 276: , [4] A. T. Brnger and W. DeLano. The direct rotation fnction: Rotational patterson correlation search applied to moleclar replacement. Acta Crystallographica, D51: , [5] M. Crisma, G. Valle, V. Monaco, F Formaggio, and C. Toniolo. n α -benzyloxycarbonylα-aminoiisobtryl-glycyl-l-isolecyl-l-lceince methyl ester monohydrate. Acta Crystallographica, C50: , [6] M. Fjinaga and R. J. Read. Experiences with a new translation-fnction program. Jornal of Applied Crystallography, 20: , [7] N. M. Glykos and M. Kokkinidis. Mltidimensional moleclar replacement. Acta Crystallographica, D57: , [8] Y. Harada, A. Lifchitz, J. Bertho, and P. Jolles. A translation fnction combining packing and diffraction information: An application to lysozyme (high-temperatre form). Acta Crystallographica, A37: , [9] D. C. Jamrog. A New Global Optimization Strategy for the Moleclar Replacement Problem. PhD thesis, Rice University, 6100 Main Street, Hoston, Texas, April Technical Report TR-0208 at [10] E. E. Lattman. Use of rotation and translation fnctions. Methods in Enzymology, 115:55 77,
13 [11] J. Navaza. Implementation of moleclar replacement in amore. Acta Crystallographica, D57: , [12] R. J. Read. Pshing the bondaries of moleclar replacement with maximm likelihood. Acta Crystallographica, D57: , [13] G. Stot and L. Jensen. X-ray Strctre Determination, A Practical Gide. John Wiley & Sons, Inc., second edition, [14] L. Tong. How to take advantage of non-crystallographic symmetry in moleclar replacement: locked rotation and translation fnctions. Acta Crystallographica, D57: ,
Modeling Roughness Effects in Open Channel Flows D.T. Souders and C.W. Hirt Flow Science, Inc.
FSI-2-TN6 Modeling Roghness Effects in Open Channel Flows D.T. Soders and C.W. Hirt Flow Science, Inc. Overview Flows along rivers, throgh pipes and irrigation channels enconter resistance that is proportional
More informationEvery manufacturer is confronted with the problem
HOW MANY PARTS TO MAKE AT ONCE FORD W. HARRIS Prodction Engineer Reprinted from Factory, The Magazine of Management, Volme 10, Nmber 2, Febrary 1913, pp. 135-136, 152 Interest on capital tied p in wages,
More informationSolutions to Assignment 10
Soltions to Assignment Math 27, Fall 22.4.8 Define T : R R by T (x) = Ax where A is a matrix with eigenvales and -2. Does there exist a basis B for R sch that the B-matrix for T is a diagonal matrix? We
More informationUsing GPU to Compute Options and Derivatives
Introdction Algorithmic Trading has created an increasing demand for high performance compting soltions within financial organizations. The actors of portfolio management and ris assessment have the obligation
More informationIn this chapter we introduce the idea that force times distance. Work and Kinetic Energy. Big Ideas 1 2 3. is force times distance.
Big Ideas 1 Work 2 Kinetic 3 Power is force times distance. energy is one-half mass times velocity sqared. is the rate at which work is done. 7 Work and Kinetic Energy The work done by this cyclist can
More informationChapter 3. 2. Consider an economy described by the following equations: Y = 5,000 G = 1,000
Chapter C evel Qestions. Imagine that the prodction of fishing lres is governed by the prodction fnction: y.7 where y represents the nmber of lres created per hor and represents the nmber of workers employed
More informationWHITE PAPER. Filter Bandwidth Definition of the WaveShaper S-series Programmable Optical Processor
WHITE PAPER Filter andwidth Definition of the WaveShaper S-series 1 Introdction The WaveShaper family of s allow creation of ser-cstomized filter profiles over the C- or L- band, providing a flexible tool
More informationSpectrum Balancing for DSL with Restrictions on Maximum Transmit PSD
Spectrm Balancing for DSL with Restrictions on Maximm Transmit PSD Driton Statovci, Tomas Nordström, and Rickard Nilsson Telecommnications Research Center Vienna (ftw.), Dona-City-Straße 1, A-1220 Vienna,
More informationLinear Programming. Non-Lecture J: Linear Programming
The greatest flood has the soonest ebb; the sorest tempest the most sdden calm; the hottest love the coldest end; and from the deepest desire oftentimes enses the deadliest hate. Socrates Th extremes of
More informationASAND: Asynchronous Slot Assignment and Neighbor Discovery Protocol for Wireless Networks
ASAND: Asynchronos Slot Assignment and Neighbor Discovery Protocol for Wireless Networks Fikret Sivrikaya, Costas Bsch, Malik Magdon-Ismail, Bülent Yener Compter Science Department, Rensselaer Polytechnic
More informationPHY2061 Enriched Physics 2 Lecture Notes Relativity 4. Relativity 4
PHY6 Enriched Physics Lectre Notes Relativity 4 Relativity 4 Disclaimer: These lectre notes are not meant to replace the corse textbook. The content may be incomplete. Some topics may be nclear. These
More informationCHAPTER ONE VECTOR GEOMETRY
CHAPTER ONE VECTOR GEOMETRY. INTRODUCTION In this chapter ectors are first introdced as geometric objects, namely as directed line segments, or arrows. The operations of addition, sbtraction, and mltiplication
More informationOn the urbanization of poverty
On the rbanization of poverty Martin Ravallion 1 Development Research Grop, World Bank 1818 H Street NW, Washington DC, USA Febrary 001; revised Jly 001 Abstract: Conditions are identified nder which the
More informationOptimal control and piecewise parametric programming
Proceedings of the Eropean Control Conference 2007 Kos, Greece, Jly 2-5, 2007 WeA07.1 Optimal control and piecewise parametric programming D. Q. Mayne, S. V. Raković and E. C. Kerrigan Abstract This paper
More information4.1 Work Done by a Constant Force
4.1 Work Done by a Constant orce work the prodct of the magnitde of an object s and the component of the applied force in the direction of the Stdying can feel like a lot of work. Imagine stdying several
More informationResource Pricing and Provisioning Strategies in Cloud Systems: A Stackelberg Game Approach
Resorce Pricing and Provisioning Strategies in Clod Systems: A Stackelberg Game Approach Valeria Cardellini, Valerio di Valerio and Francesco Lo Presti Talk Otline Backgrond and Motivation Provisioning
More informationOptimal Personalized Filtering Against Spear-Phishing Attacks
Optimal Personalized Filtering Against Spear-Phishing Attacks Aron Laszka and Yevgeniy Vorobeychik and Xenofon Kotsokos Institte for Software Integrated Systems Department of Electrical Engineering and
More informationEquilibrium of Forces Acting at a Point
Eqilibrim of orces Acting at a Point Eqilibrim of orces Acting at a Point Pre-lab Qestions 1. What is the definition of eqilibrim? Can an object be moving and still be in eqilibrim? Explain.. or this lab,
More informationHorses and Rabbits? Optimal Dynamic Capital Structure from Shareholder and Manager Perspectives
Horses and Rabbits? Optimal Dynamic Capital trctre from hareholder and Manager Perspectives Nengji J University of Maryland Robert Parrino University of exas at Astin Allen M. Poteshman University of Illinois
More informationMarket Impact and Optimal Equity Trade Scheduling
Market Impact and Optimal Eqity Trade Schedling Dan dibartolomeo Northfield Information Services, Inc. Colmbia University Math Finance Seminar September 2007 Presentation Otline Brief review of the optimal
More informationFUNCTIONAL COEFFICIENT MODELS UNDER UNIT ROOT BEHAVIOR. 1. Introduction
FUNCTIONAL COEFFICIENT MODELS UNDER UNIT ROOT BEHAVIOR TED JUHL Abstract. We analyze the statistical properties of nonparametrically estimated fnctions in a fnctional-coefficient model if the data has
More informationA Spare Part Inventory Management Model for Better Maintenance of Intelligent Transportation Systems
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 A Spare Part Inventory Management Model for Better Maintenance of Intelligent
More informationPrimary Analysis of Effective Permeability of the Flame in Burning Natural Gas
Jornal of etals, aterials and inerals. Vol.7 No. pp.63-66. rimary Analysis of Effective ermeability of the Flame in Brning Natral Gas Rakoš JAROSAV * and Repasova AGDAENA * Department of Thermal Technology,
More informationResearch on Pricing Policy of E-business Supply Chain Based on Bertrand and Stackelberg Game
International Jornal of Grid and Distribted Compting Vol. 9, No. 5 (06), pp.-0 http://dx.doi.org/0.457/ijgdc.06.9.5.8 Research on Pricing Policy of E-bsiness Spply Chain Based on Bertrand and Stackelberg
More informationRegular Specifications of Resource Requirements for Embedded Control Software
Reglar Specifications of Resorce Reqirements for Embedded Control Software Rajeev Alr and Gera Weiss University of Pennsylvania Abstract For embedded control systems a schedle for the allocation of resorces
More informationeconstor zbw www.econstor.eu
econstor www.econstor.e Der Open-Access-Pblikationsserver der ZBW Leibniz-Informationszentrm Wirtschaft The Open Access Pblication Server of the ZBW Leibniz Information Centre for Economics MacMinn, Richard;
More informationNewton s three laws of motion, the foundation of classical. Applications of Newton s Laws. Chapter 5. 5.1 Equilibrium of a Particle
Chapter 5 Applications of Newton s Laws The soles of hiking shoes are designed to stick, not slip, on rocky srfaces. In this chapter we ll learn abot the interactions that give good traction. By the end
More informationOptimal Trust Network Analysis with Subjective Logic
The Second International Conference on Emerging Secrity Information, Systems and Technologies Optimal Trst Network Analysis with Sbjective Logic Adn Jøsang UNIK Gradate Center, University of Oslo Norway
More informationCosmological Origin of Gravitational Constant
Apeiron, Vol. 5, No. 4, October 8 465 Cosmological Origin of Gravitational Constant Maciej Rybicki Sas-Zbrzyckiego 8/7 3-6 Krakow, oland rybicki@skr.pl The base nits contribting to gravitational constant
More informationSample Pages. Edgar Dietrich, Alfred Schulze. Measurement Process Qualification
Sample Pages Edgar Dietrich, Alfred Schlze Measrement Process Qalification Gage Acceptance and Measrement Uncertainty According to Crrent Standards ISBN: 978-3-446-4407-4 For frther information and order
More informationEffect of flow field on open channel flow properties using numerical investigation and experimental comparison
INTERNATIONAL JOURNAL OF ENERGY AND ENVIRONMENT Volme 3, Isse 4, 2012 pp.617-628 Jornal homepage: www.ijee.ieefondation.org Effect of flow field on open channel flow properties sing nmerical investigation
More informationTrustSVD: Collaborative Filtering with Both the Explicit and Implicit Influence of User Trust and of Item Ratings
TrstSVD: Collaborative Filtering with Both the Explicit and Implicit Inflence of User Trst and of Item Ratings Gibing Go Jie Zhang Neil Yorke-Smith School of Compter Engineering Nanyang Technological University
More informationIntroduction to HBase Schema Design
Introdction to HBase Schema Design Amandeep Khrana Amandeep Khrana is a Soltions Architect at Clodera and works on bilding soltions sing the Hadoop stack. He is also a co-athor of HBase in Action. Prior
More informationStability of Linear Control System
Stabilit of Linear Control Sstem Concept of Stabilit Closed-loop feedback sstem is either stable or nstable. This tpe of characterization is referred to as absolte stabilit. Given that the sstem is stable,
More information11 Success of the Help Desk: Assessing Outcomes
11 Sccess of the Help Desk: Assessing Otcomes I dread sccess... I like a state of continal becoming, with a goal in front and not behind. George Bernard Shaw Key Findings Respondents help desks tend to
More information3. Fluid Dynamics. 3.1 Uniform Flow, Steady Flow
3. Flid Dynamics Objectives Introdce concepts necessary to analyse flids in motion Identify differences between Steady/nsteady niform/non-niform compressible/incompressible flow Demonstrate streamlines
More informationHOMOTOPY FIBER PRODUCTS OF HOMOTOPY THEORIES
HOMOTOPY FIBER PRODUCTS OF HOMOTOPY THEORIES JULIA E. BERGNER Abstract. Gien an appropriate diagram of left Qillen fnctors between model categories, one can define a notion of homotopy fiber prodct, bt
More informationSingle-Year and Multi-Year Insurance Policies in a Competitive Market
Single-Year and Mlti-Year Insrance Policies in a Competitive Market Pal R. Kleindorfer INSEAD, France Howard Knrether The Wharton School University of Pennsylvania Chieh O-Yang City University of Hong
More informationIn the insurance business risky investments are dangerous
Noname manscript No. (will be inserted by the editor) In the insrance bsiness risy investments are dangeros Anna Frolova 1, Yri Kabanov 2, Sergei Pergamenshchiov 3 1 Alfaban, Moscow, Rssia 2 Laboratoire
More informationA priori error analysis of stabilized mixed finite element method for reaction-diffusion optimal control problems
F et al. Bondary Vale Problems 2016 2016:23 DOI 10.1186/s13661-016-0531-9 R E S E A R C H Open Access A priori error analysis of stabilized mixed finite element metod for reaction-diffsion optimal control
More informationOPTIMIZATION AND CONTROL
5.5 Optimal stopping over the infinite horizon................. 19 Contents OPTIMIZATION AND CONTROL Richard Weber DYNAMIC PROGRAMMING 1 1 Dynamic Programming: The Optimality Eqation 1 1.1 Control as optimization
More information10 Evaluating the Help Desk
10 Evalating the Help Desk The tre measre of any society is not what it knows bt what it does with what it knows. Warren Bennis Key Findings Help desk metrics having to do with demand and with problem
More informationKey words. renewal risk models, surplus dependent premiums, boundary value problems, Green s operators, asymptotic expansions,
EXACT AND ASYMPTOTIC RESULTS FOR INSURANCE RISK MODELS WITH SURPLUS-DEPENDENT PREMIUMS HANSJÖRG ALBRECHER, CORINA CONSTANTINESCU, ZBIGNIEW PALMOWSKI, GEORG REGENSBURGER, AND MARKUS ROSENKRANZ Abstract.
More information9 Setting a Course: Goals for the Help Desk
IT Help Desk in Higher Edcation ECAR Research Stdy 8, 2007 9 Setting a Corse: Goals for the Help Desk First say to yorself what yo wold be; and then do what yo have to do. Epictets Key Findings Majorities
More informationCIVE2400 Fluid Mechanics. Section 1: Fluid Flow in Pipes
CIVE00 Flid Mechanics Section : Flid Flow in Pipes CIVE00 FLUID MECHNICS... SECTION : FLUID FLOW IN PIPES.... FLUID FLOW IN PIPES.... Pressre loss de to riction in a pipeline..... Pressre loss dring laminar
More informationBLIND speech separation (BSS) aims to recover source
A Convex Speech Extraction Model and Fast Comptation by the Split Bregman Method Meng Y, Wenye Ma, Jack Xin, and Stanley Osher. Abstract A fast speech extraction (FSE) method is presented sing convex optimization
More informationEnergy exchanges during indentation tests. in fresh-water ice
Annals of Glaciology 15 1991 < nternational Glaciological Society nergy exchanges dring indentation tests in fresh-water ice DVNDR S. SODH U.S. Army Cold Regions Research and ngineering Laboratory, 7 Lyme
More informationImmigration and Unemployment of Skilled and Unskilled Labor. Shigemi Yabuuchi* Nagoya City University
Ag. 7, 6 Immigration and Unemployment of Skilled and Unskilled Labor Shigemi Yabchi* Nagoya City University Abstract This paper discsses the problem of nemployment in developed contries that faces international
More informationA Contemporary Approach
BORICP01.doc - 1 Second Edition Edcational Psychology A Contemporary Approach Gary D. Borich The University of Texas at Astin Martin L. Tombari University of Denver (This pblication may be reprodced for
More informationChapter 2. ( Vasiliy Koval/Fotolia)
hapter ( Vasili Koval/otolia) This electric transmission tower is stabilied b cables that eert forces on the tower at their points of connection. In this chapter we will show how to epress these forces
More informationFaster Inversion and Other Black Box Matrix Computations Using Efficient Block Projections
Faster Inversion and Other Black Box Matrix Comptations Using Efficient Block Projections Wayne Eberly 1, Mark Giesbrecht, Pascal Giorgi,, Arne Storjohann, Gilles Villard (1) Department of Compter Science,
More informationHow To Plan A Cloud Infrastructure
Concrrent Placement, Capacity Provisioning, and Reqest Flow Control for a Distribted Clod Infrastrctre Shang Chen, Yanzhi Wang, Massod Pedram Department of Electrical Engineering University of Sothern
More informationOn a Generalized Graph Coloring/Batch Scheduling Problem
Reglar Papers On a Generalized Graph Coloring/Batch Schedling Problem Giorgio Lcarelli 1, Ioannis Milis Dept. of Informatics, Athens University of Economics and Bsiness, 104 34, Athens, Greece, {glc, milis}@aeb.gr
More informationNAZIA KANWAL VECTOR TRACKING LOOP DESIGN FOR DEGRADED SIGNAL ENVIRONMENT. Master of Science Thesis
NAZIA KANWAL VECTOR TRACKING LOOP DESIGN FOR DEGRADED SIGNAL ENVIRONMENT Master of Science Thesis Examiners: Professor Jari Nrmi, Adjnct Professor Simona Lohan and Dr. Heikki Hrskainen Examiner and topic
More informationData De-duplication from the data sets using Similarity functions
Data De-dplication from the data sets sing Similarity fnctions M.Chitrarpa 1, V.Mniraj Naid 2 1 M.Tech Stdent Department of CSE Adisankara College of Engineering & Technology Gdr, Nellore district, India
More informationCloser Look at ACOs. Making the Most of Accountable Care Organizations (ACOs): What Advocates Need to Know
Closer Look at ACOs A series of briefs designed to help advocates nderstand the basics of Accontable Care Organizations (ACOs) and their potential for improving patient care. From Families USA Updated
More informationOPTIMAL TRADE CREDIT REINSURANCE PROGRAMS WITH SOLVENCY REQUIREMENTS
OPTIMAL TRADE CREDIT REINSURANCE PROGRAMS WITH SOLVENCY REQUIREMENTS Lca Passalacqa Dipartimento di Scienze Attariali e Finanziarie La Sapienza, Università di Roma Viale Regina Elena, 295 00161 Roma, Italy
More informationPhase determination methods in macromolecular X- ray Crystallography
Phase determination methods in macromolecular X- ray Crystallography Importance of protein structure determination: Proteins are the life machinery and are very essential for the various functions in the
More informationCompensation Approaches for Far-field Speaker Identification
Compensation Approaches for Far-field Speaer Identification Qin Jin, Kshitiz Kmar, Tanja Schltz, and Richard Stern Carnegie Mellon University, USA {qjin,shitiz,tanja,rms}@cs.cm.ed Abstract While speaer
More informationEvolutionary Path Planning for Robot Assisted Part Handling in Sheet Metal Bending
Evoltionary Path Planning for Robot Assisted Part Handling in Sheet Metal Bending Abstract Xiaoyn Liao G. Gary Wang * Dept. of Mechanical & Indstrial Engineering, The University of Manitoba Winnipeg, MB,
More informationGUIDELINE. Guideline for the Selection of Engineering Services
GUIDELINE Gideline for the Selection of Engineering Services 1998 Mission Statement: To govern the engineering profession while enhancing engineering practice and enhancing engineering cltre Pblished by
More informationA Novel QR Code and mobile phone based Authentication protocol via Bluetooth Sha Liu *1, Shuhua Zhu 2
International Conference on Materials Engineering and Information Technology Applications (MEITA 2015) A Novel QR Code and mobile phone based Athentication protocol via Bletooth Sha Li *1, Shha Zh 2 *1
More informationCandidate: Suzanne Maxwell. Date: 09/19/2012
Medical Coder / Billing Clerk Assessment Report Szanne Maxwell 09/19/2012 www.resorceassociates.com Szanne Maxwell 09/19/2012 Prepared For: NAME Prepared by: John Lonsbry, Ph.D. & Lcy Gibson, Ph.D., Licensed
More informationCorporate performance: What do investors want to know? Innovate your way to clearer financial reporting
www.pwc.com Corporate performance: What do investors want to know? Innovate yor way to clearer financial reporting October 2014 PwC I Innovate yor way to clearer financial reporting t 1 Contents Introdction
More informationBonds with Embedded Options and Options on Bonds
FIXED-INCOME SECURITIES Chapter 14 Bonds with Embedded Options and Options on Bonds Callable and Ptable Bonds Instittional Aspects Valation Convertible Bonds Instittional Aspects Valation Options on Bonds
More informationCSC 505, Fall 2000: Week 8
Objecties: CSC 505, Fall 2000: Week 8 learn abot the basic depth-first search algorithm learn how properties of a graph can be inferred from the strctre of a DFS tree learn abot one nontriial application
More informationFINANCIAL FITNESS SELECTING A CREDIT CARD. Fact Sheet
FINANCIAL FITNESS Fact Sheet Janary 1998 FL/FF-02 SELECTING A CREDIT CARD Liz Gorham, Ph.D., AFC Assistant Professor and Family Resorce Management Specialist, Utah State University Marsha A. Goetting,
More informationI Symbolization J,1 II e L~ "-"-:"u"'dll... Table I: The kinds of CGs and their classification, (where, t - a local neighbourhood topology)
POSTER SESSIONS 484 REPRESENTATION OF THE GENERALIZED DATA STRUCTURES FOR MULTI-SCALE GIS M.O.Govorov Dept. of Cartography,' Siberian State Academy of Geodesy Plahotnogo 10, Novosibirsk, 630108, Rssia
More informationCell Radius Inaccuracy: A New Measure of Coverage Reliability
Cell adis Inaccracy: A ew Measre of Coverage eliaility PETE BEADI, MEMBE, IEEE, MEG YEE, MEMBE, IEEE, AD THOMAS ELLIS Preprint of Plication in the ovemer 1998 Isse (Vol. 47, o.4, pp. 115-16) of the IEEE
More information8. Forced Convection Heat Transfer
8. Forced Convection Heat Transfer 8.1 Introdction The general definition for convection ma be smmarized to this definition "energ transfer between the srface and flid de to temperatre difference" and
More informationMake the College Connection
Make the College Connection A college planning gide for stdents and their parents Table of contents The compelling case for college 2 Selecting a college 3 Paying for college 5 Tips for meeting college
More informationCloser Look at ACOs. Putting the Accountability in Accountable Care Organizations: Payment and Quality Measurements. Introduction
Closer Look at ACOs A series of briefs designed to help advocates nderstand the basics of Accontable Care Organizations (ACOs) and their potential for improving patient care. From Families USA Janary 2012
More informationLIMITS IN CATEGORY THEORY
LIMITS IN CATEGORY THEORY SCOTT MESSICK Abstract. I will start assming no knowledge o category theory and introdce all concepts necessary to embark on a discssion o limits. I will conclde with two big
More informationCentral Angles, Arc Length, and Sector Area
CHAPTER 5 A Central Angles, Arc Length, and Sector Area c GOAL Identify central angles and determine arc length and sector area formed by a central angle. Yo will need a calclator a compass a protractor
More informationDesigning and Deploying File Servers
C H A P T E R 2 Designing and Deploying File Servers File servers rnning the Microsoft Windows Server 2003 operating system are ideal for providing access to files for sers in medim and large organizations.
More informationStock Market Liquidity and Macro-Liquidity Shocks: Evidence from the 2007-2009 Financial Crisis
Stock Market Liqidity and Macro-Liqidity Shocks: Evidence from the 2007-2009 Financial Crisis Chris Florackis *, Alexandros Kontonikas and Alexandros Kostakis Abstract We develop an empirical framework
More informationTowers Watson Manager Research
Towers Watson Manager Research How we se fnd performance data Harald Eggerstedt 13. März 212 212 Towers Watson. All rights reserved. Manager selection at Towers Watson The goal is to find managers that
More informationDesigning an Authentication Strategy
C H A P T E R 1 4 Designing an Athentication Strategy Most organizations need to spport seamless access to the network for mltiple types of sers, sch as workers in offices, employees who are traveling,
More informationThe Intelligent Choice for Disability Income Protection
The Intelligent Choice for Disability Income Protection provider Pls Keeping Income strong We prposeflly engineer or disability income prodct with featres that deliver benefits sooner and contine paying
More informationManipulating Deformable Linear Objects: Characteristic Features for Vision-Based Detection of Contact State Transitions
Maniplating Deformable Linear Objects: Characteristic Featres for Vision-Based Detection of Contact State Transitions Jürgen Acker Dominik Henrich Embedded Systems and Robotics Lab. (RESY) Faclty of Informatics,
More informationPerforming Arithmetic Operations on Round-to-Nearest Representations
1 Performing Arithmetic Operations on Rond-to-Nearest Representations Peter Kornerp, Member, IEEE Jean-Michel Mller, Senior Member, IEEE Adrien Panhalex Abstract Dring any composite comptation there is
More informationEffect of Angular Velocity of Inner Cylinder on Laminar Flow through Eccentric Annular Cross Section Pipe
Asian Transactions on Engineering (ATE ISSN: -467) Volme 3 Isse Effect of Anglar Velocity of Inner Cylinder on Laminar Flow throgh Eccentric Annlar Cross Section Pipe Ressan Faris Hamd * Department of
More informationSickness Absence in the UK: 1984-2002
Sickness Absence in the UK: 1984-2002 Tim Barmby (Universy of Drham) Marco Ecolani (Universy of Birmingham) John Treble (Universy of Wales Swansea) Paper prepared for presentation at The Economic Concil
More informationAn unbiased crawling strategy for directed social networks
Abstract An nbiased crawling strategy for directed social networks Xeha Yang 1,2, HongbinLi 2* 1 School of Software, Shenyang Normal University, Shenyang 110034, Liaoning, China 2 Shenyang Institte of
More informationPurposefully Engineered High-Performing Income Protection
The Intelligent Choice for Disability Income Insrance Prposeflly Engineered High-Performing Income Protection Keeping Income strong We engineer or disability income prodcts with featres that deliver benefits
More informationAn accelerated Branch-and-Bound algorithm for assignment problems of utility systems
Compters and Chemical Engineering 6 (00) 67 630 www.elsevier.com/locate/compchemeng An accelerated Branch-and-Bond algorithm for assignment problems of tility systems Alexandros M. Strovalis a, Istvan
More informationDIRECT TAX LAWS Taxability of Capital Gains on By-back of Shares - Debate ignites after AAR s rling in RST s case BACKGROUND 1. Recently, the Athority for Advance Rlings ( AAR ) in the case of RST, In
More informationCandidate: Shawn Mullane. Date: 04/02/2012
Shipping and Receiving Specialist / Inventory Control Assessment Report Shawn Mllane 04/02/2012 www.resorceassociates.com To Improve Prodctivity Throgh People. Shawn Mllane 04/02/2012 Prepared For: NAME
More informationEffective governance to support medical revalidation
Effective governance to spport medical revalidation A handbook for boards and governing bodies This docment sets ot a view of the core elements of effective local governance of the systems that spport
More informationRoth 401(k) and Roth 403(b) Accounts: Pay Me Now or Pay Me Later Why a Roth Election Should Be Part of Your Plan Now
Reprinted with permission from the Society of FSP. Reprodction prohibited withot pblisher's written permission. Roth 401(k) and Roth 403(b) Acconts: Why a Roth Election Shold Be Part of Yor Plan Now by
More information3 Distance in Graphs. Brief outline of this lecture
Distance in Graphs While the preios lectre stdied jst the connectiity properties of a graph, now we are going to inestigate how long (short, actally) a connection in a graph is. This natrally leads to
More informationUNIT 62: STRENGTHS OF MATERIALS Unit code: K/601/1409 QCF level: 5 Credit value: 15 OUTCOME 2 - TUTORIAL 3
UNIT 6: STRNGTHS O MTRIS Unit code: K/601/1409 QC level: 5 Credit vale: 15 OUTCOM - TUTORI 3 INTRMDIT ND SHORT COMPRSSION MMBRS Be able to determine the behavioral characteristics of loaded beams, colmns
More informationCloser Look at ACOs. Designing Consumer-Friendly Beneficiary Assignment and Notification Processes for Accountable Care Organizations
Closer Look at ACOs A series of briefs designed to help advocates nderstand the basics of Accontable Care Organizations (ACOs) and their potential for improving patient care. From Families USA Janary 2012
More informationStock Market Liquidity and Macro-Liquidity Shocks: Evidence from the 2007-2009 Financial Crisis
Stock Market Liqidity and Macro-Liqidity Shocks: Evidence from the 2007-2009 Financial Crisis Chris Florackis *, Alexandros Kontonikas and Alexandros Kostakis Abstract We develop an empirical framework
More informationLimited liability, contractual choice, and the tenancy ladder
Jornal of Development Economics Vol. 66 200 289 303 www.elsevier.comrlocatereconbase Limited liability, contractal choice, and the tenancy ladder Tridip Ray a, Nirvikar Singh b,) a ong Kong UniÕersity
More informationA Stdy on Cstomer Service Qality of Banks in India Dr. Manasa Nagabhshanam Lead Researcher Analyz Research Soltions Pvt. Ltd. Bangalore BLANK Table of Contents Chapter 1 315-317 Introdction 315 1.1 Role
More informationAnalytical Calculation of Risk Measures for Variable Annuity Guaranteed Benefits
Analytical Calclation of Risk Measres for Variable Annity Garanteed Benefits Rnhan Feng Department of Mathematical Sciences University of Wisconsin - Milwakee fengr@wm.ed Hans W. Volkmer Department of
More informationMSc and MA in Finance and Investment online Study an online MSc and MA in Finance and Investment awarded by UNINETTUNO and Geneva Business School
MSc and MA in Finance and Investment online Stdy an online awarded by UNINETTUNO and Geneva Bsiness School Awarded by Geneva Bsiness School Geneva Barcelona Moscow Class profile The connects yo with stdents
More information1.3. DOT PRODUCT 19. 6. If θ is the angle (between 0 and π) between two non-zero vectors u and v,
1.3. DOT PRODUCT 19 1.3 Dot Product 1.3.1 Definitions and Properties The dot product is the first way to multiply two vectors. The definition we will give below may appear arbitrary. But it is not. It
More information