### Table 1 Sample Statistics US Semiconductor Sample before Cleaning 1525 Observations (110 Firms) 1965-1997 Variable Name Mean S.D. Median First Q Third Q Min Max

in The Patent Paradox Revisited: Determinants of Patenting in the US Semiconductor Industry, 1980-94

1999

"... In PAGE 17: ...years), and to determine whether the firm owned and operated its own fab (manufacturer) or specialized in product design alone (design firm).37 Table1 gives some summary statistics for our key variables; the top panel is for our universe of firms from 1965 to 1997 and the bottom panel is based on the sample we use for estimation. The median firm in our sample is 27 years old, has approximately 500 employees, does $5.... ..."

Cited by 7

### Table I. MIMO Transceiver Structures - Downstream SVD ZF NSP

### Table II. MIMO Transceiver Structures - Upstream SVD ZF NSP

### Table I. MIMO Transceiver Structures - Downstream SVD ZF NSP

### Table II. MIMO Transceiver Structures - Upstream SVD ZF NSP

### Table 2 Geometrically conforming domain decomposition; Matching grids across the interface.

1999

"... In PAGE 6: ...reconditioner M, cf. (9), and the Dirichlet preconditioner M, cf. (6). As a comparison, we also present iteration counts for the geometrically conforming case using the preconditioners c M and M, cf. Table2 . Here, is partitioned in a... ..."

Cited by 3

### Table 2. Summary soil pH and Electrical Conductivity statistics.

"... In PAGE 6: ...81). Site level examination of the pH values for soil samples taken next to ant mounds as well as loca- tions where they did not occur show a nearly identi- cal pattern ( Table2 ). There was no significant difference between the geometric means of the two sets of samples; soils with ant mounds averaging 7.... ..."

### Table 1: Geometric information

"... In PAGE 20: ... e = QG gt;(G gt;QG) 1e fforcing the constraint G gt; e = eg x01 = 0; x02 = 0; x03 = e fchoose the initial guessg b1 = g; b2 = d = f B gt; e; b3 = 0 fcompute right hand sideg d = K3x0 b fcompute the defectg r0 = D3 ^ L 1 3 d fapply the transformationg p0 = w0 = D 1 3 r0 fpreconditioningg 0 = (w0; r0) for n = 0 step 1 until n quot; 0 do vn = D3 ^ L 1 3 K3pn fmatrix vector multiplication and transformationg = (vn; pn) = n+1= xn+1 = xn pn fupdate the iterateg rn+1 = rn vn fupdate the defectg wn+1 = D 1 3 rn+1 fpreconditioningg n+1 = (wn+1; rn+1) = n+1= n pn+1 = wn+1 + pn fupdate of the search directiong end for Figure 1: Domain decomposition with 8 subdomains In Table1 , the geometric informations about the domain decomposition and the discretization are listed for the re nement levels L used. Starting from the coarsest grid with 192 triangles for the whole domain , the re ned meshes are recursively constructed by subdividing each triangle into four smaller similar triangles.... In PAGE 21: ...Table 1: Geometric information In all following tables, a uni ed notation is used. L again denotes the re nement level and Table1 gives the corresponding information of the grids. t1 and t2 are the measured times in seconds for setting up the corresponding system of linear equations and for their solution.... ..."

Cited by 1

### Table 2 presents the geometric mean of the IPC and

"... In PAGE 3: ... The IPC results for a perfect cache| a cache that never generates a miss#7B will be used for compar- ison. Table2 : Geometric mean of IPC #28and percentage of perfect IPC#29 for the three schemes.... ..."

### Table 3: Geometric means of the comparative results of LANCELOT

"... In PAGE 15: ... Problems CHENHARK (number 3), JNLBRNG1 (number 5) and JNLBRNGB (number 8) are more favorable to the nonmonotone strategy, whereas the opposite happens to problems NCVXBQP2 (number 10), OBSTCLAL (number 14) and OBSTCLBU (number 17), for which the monotone algorithm performs better. In Figures 2 and 3 we can visualize the comparative results between the nonmonotone algorithm and the combination that performed best for LANCELOT according to Table3 , namely, using preconditioned conjugate gra- dient and computing the exact Cauchy point (PCGEX). We plot the loga- rithms, to the base 10, of the ratios between the results of the nonmonotone algorithm and LANCELOT, analyzing, in Figure 2, the number of iterations performed, and, in Figure 3, the CPU time spent.... ..."