### Table 2 Euclidean norms quot; of matrix inversion calculations using double-precision arithmetic for matrix inner products.

"... In PAGE 3: ... (24); 5) single-pass 3rd-order; and 6) direct. The Euclidean norm values shown in Table2 are gen- erally smaller than those in Table 1 as a result of using double-precision arithmetic for the inner products in the matrix multiplications. These results are also more representative of the true error in each method since numerical errors are minimized.... In PAGE 3: ... Using double- precision arithmetic, however, the higher-order iterative corrections did improve the 2 X 2 matrix calculation. One can observe from Table2 that the direct method is about as accurate as all other methods for the test ma- trices larger than 2 X 2. The second series of tests was made to confirm this result and to check that the actual inversion errors being studied are free of unrelated numerical errors.... ..."

### Table 1. Summary of reactions between sodium and organochlorosilanes Product

"... In PAGE 1: ... EXPERIMENTAL Materials Me2SiCl2,MeSiCl3,MeHSiCl2, and NaH purchased from Aldrich and PhSiCl3 obtained from Echo Chemical were used in the reactions without further purification. Synthesis of sample DD from Me2SiCl2 and Na A summary of reaction conditions and experimental observations is listed in Table1 . The reaction employing Me2SiCl2 in reaction with Na is described below as a typi- cal example.... In PAGE 2: ... They were processed directly into the final products PT- 573 and PT-723. A summary of the experimental data, in- cluding the overall reaction conditions and general infor- mation of the products is listed in Table1 . These will be 1478 J.... ..."

### Table 1 Performance of the TRL eigenvalue calculations using hierarchical matrix approximation for fast matrix-vector products

2005

### Table 4.2 shows which sectors gained shares and which lost most heavily in the two countries. Having said this, the correlation is by no means perfect. For instance, Germany, Austria and Belgium all grew relatively rapidly, while their structural change was slow. Contrarily, there was substantial structural change in Sweden and Finland, but their growth was relatively slow.

"... In PAGE 17: ... The creation of these integrated enterprise networks has far-reaching effects on European restructuring and integration. Table4 : Evolution of MNE strategies and structures Form Types of intra- firm linkages Degree of integration Environment Stand-alone Ownership, technology Weak Host country accessible to FDI; significant trade barriers; costly communications and transportation Simple integration Ownership, technology, markets, finance, other inputs Partially strong Bilaterally open trade and FDI; non-equity arrangements Complex international production All functions Potentially strong overall Open trade and FDI; IT; convergence in tastes; increased competition Source: World Investment Report 1993 (United Nations). First, the networking of firms is essential for the cross- border transfer of know-how and of proprietary advantages.... ..."

### Table 4.1: Growth of production, employment and exports and the speed of change Rank correlation coefficient between

"... In PAGE 17: ... The creation of these integrated enterprise networks has far-reaching effects on European restructuring and integration. Table4 : Evolution of MNE strategies and structures Form Types of intra- firm linkages Degree of integration Environment Stand-alone Ownership, technology Weak Host country accessible to FDI; significant trade barriers; costly communications and transportation Simple integration Ownership, technology, markets, finance, other inputs Partially strong Bilaterally open trade and FDI; non-equity arrangements Complex international production All functions Potentially strong overall Open trade and FDI; IT; convergence in tastes; increased competition Source: World Investment Report 1993 (United Nations). First, the networking of firms is essential for the cross- border transfer of know-how and of proprietary advantages.... ..."

### Table 5. We have calculated the improvement of MCF-CA over the other two schemes in Table 6. In Table 7 indicates that our scheme results in higher network utilization than the two other schemes.

2003

"... In PAGE 11: ...9 1 0 2000 4000 6000 8000 10000 Requested BW B l o c k i n g R a t i o SPF-CA LCP-CA MCF-CA Figure 7: Partial result for Experiment 1 on topology 2. SPF-CA LCP-CA MCF-CA Topology1 1228 1419 1467 Topology2 790 837 856 Table5 : The total amount of accepted bandwidth at the saturation point for Experiment 1. SPF-CA LCP-CA Topology1 19.... ..."

Cited by 4

### Table 2. Structure Matching DataType and ProductType ProductType

"... In PAGE 13: ...4) Let us now apply the algorithm discussed above to match the complex data types DataType and ProductType of the two WSDL speci cations shown in Figure 1. Because both data types are complex, the algorithm needs to recursively match all the elements of the two complex structures to decide on their similarity scores, thus a 3 4 matrix is constructed ( Table2 ). We use the notation ? ! to indicate this recursive process; the question mark indicates that a match score is currently unavailable and it will eventually be replaced by a match score obtained from fur- ther recursive calculations.... In PAGE 13: ...Table 3. Structure Matching Item and ProductParts ProductParts Item Part: String Quantity: int 5 Component: String 10 Table2 shows how the complex data structures DataType and ProductType are matched. Primitive data types are compared by a simple table look up; if either data type being compared is composite, their match score cannot be calculated till their constituent elements are matched.... In PAGE 14: ... Thus, DataType ! Item maps to ProductType ! ProductParts with a score of 20. The bottom-right cell of Table2 , which corresponds to this match, now has the value of ? ! 20. The rest of the Table 2 cells are similarly lled and then the matching score be- tween DataType and ProductType can be determined.... In PAGE 14: ... The bottom-right cell of Table 2, which corresponds to this match, now has the value of ? ! 20. The rest of the Table2 cells are similarly lled and then the matching score be- tween DataType and ProductType can be determined. The algorithm exhaustively forms all possible pair-wise combinations of DataType and ProductType elements.... In PAGE 14: ... Table 4 lists the four best matches with scores of 35 between DataType and ProductType. Elements of DataType in column 2 are matched to elements of ProductType in col- umn 3, and the scores in column 4 are their corresponding match scores according to Table2 . Finally DataType and ProductType have the same grouping style of lt;all gt;, and a bonus score of 10 is added; as a result, the query service DataType matches to the target service data type ProductType with a match score of 45.... ..."

### Table 2: Density Improvement for Various Coding Schemes

"... In PAGE 7: ... When used with a di erential encoder, these constraints work to prevent isi from causing erroneous demodulation. Comparing the coded systems with uncoded systems yields the density im- provements in Table2 . These numbers re ect data density that has been adjusted by code rate and assume the existence of encoders and decoders that can come arbitrarily close to capacity values calculated by Weeks and Blahut [18].... In PAGE 7: ... Designing capacity-achieving encoders, decoders, and appropriate di eren- tial encoders remains an open problem. Because such encoders and decoders have not been designed, the density improvement values in Table2 provide only a theoretical estimate.... In PAGE 8: ...The numbers in Table2 show that as the extent of the isolation of ones gets larger, greater density improvement is possible. Correspondingly, larger constraints require greater compu- tational complexity to implement encoders and decoders.... ..."

### Table 2: Access times for a 4K-entry gshare predictor vs. the Boolean formula predictor. The delays were obtained us- ing an HSPICE model for the Boolean formula predictor and ECacti for gshare.

2001

"... In PAGE 5: ... This circuit makes a branch prediction based on a history length of 8 and an 8-bit encoding of a read-once Boolean formula. surements are the time from the midpoint of the input sig- nal switching to the midpoint of the output signal switching; these delays are shown in Table2 for C6 BP BK and C6 BP BDBI inputs. We calculated the lookup time for a gshare predictor using the ECacti tool.... In PAGE 5: ... We calculated the lookup time for a gshare predictor using the ECacti tool. Table2 shows the access times for a 4K-entry gshare predictor and two sizes of the Boolean for- mula predictor, C6 BP BK and C6 BP BDBI, for a range of fabrica- tion technologies. We chose the 4K-entry predictor because, as we will see in Section 4, the C6 BP BK version of the Boolean formula predictor only slightly exceeds the accuracy of a 4K- entry gshare.... In PAGE 5: ... As fabrication technology improves, transistors can be made smaller and faster, resulting in higher clock frequencies and faster combinational circuits. As Table2 shows, access times for each structure improve as the minimum feature size decreases. The Boolean formula predictor is consistently faster than the 4K-entry gshare predictor, allowing more time for com- munication and computation within a clock cycle.... ..."

Cited by 10

### Table 1 summarizes communication and computational complexities associated with the aggre- gation schemes, as a function of the number of border nodes, b. All the aggregation schemes are computed on the latest reported topology induced by the border nodes, no averaging was used. The are several reasons not to use the full network topology in calculating the aggregation. The most obvious one is that this way it is easy to encode the aggregation in the PNNI standard using ex- ceptions [AS98]. When security is of concern, this way no details of the internal network structure are revealed. Finally, for aggregation schemes with high calculation complexity, the use of a smaller topology reduces the calculation overhead. It is important to note that aggregation calculation does not come for free, and the calculation cost of an aggregation scheme must be weigh against the advantage of its accuracy. The performance of the Complete scheme serves as a control variable in our experiments, and is used to assess the loss in network performance due to aggregation. In some cases, especially when the link delay is low, Complete exhibits slightly worse performance than other schemes. This fact, perhaps surprising at rst, is due to the on-line nature of the problem. Even an \optimal quot; (alas near-sighted) decision made by the omniscient Complete algorithm may prove to be very bad in

"... In PAGE 7: ...representation calculation scheme size complexity Complete b(b ? 1) - DIA/AVE 1 O(b2) MST b ? 1 O(b2) RST b ? 1 O(b2) Spanner O(b1+1=t) O(b4) Table1 : A summary of the aggregation schemes simulated in this paper terms of the success of future (yet unknown) connection requests. 2.... ..."