Half-interval

Through the approximation of every heat conduction equation with number a on the half-interval ij+(a-i)/p < t < ij+ajp by the standard two-layer weighted scheme we arrive at the chain of p one-dimensional schemes... 

For convenience, only four stages were used in this model. An iterative solution is required for the bubble point calculations and this is based on the half-interval method. A FORTRAN subroutine EQUIL, incorporated in the ISIM program, estimates the equilibrium conditions for each plate. The iteration routine was taken from Luyben and Wenzel (1988). The program runs very slowly. 

If LINAIOD is zero, the model is treated as nonlinear, and a test is made for convergence of the parameter estimation. Convergence is normally declared, and final statistics of Section 6.6 are reported, if no free parameter is at a CHMAX limit and if j for each basis parameter is less than one-tenth of the 95% posterior probability half-interval calculated from Eq. (6.6-10). If the number of basis parameters equals the number of events, leaving no degrees of freedom for the estimation, this interval criterion is replaced by Aef < RPTOL... 

The case KEY= 2 in the routine QXG in FLR [5] uses a family of four formulas in the RMS formulas, namely, 13-point (L), 19-point h), 27-point (A), 41-point (D rules. Including Ninomiya s stable 11-point rule (7o) yields a sequence of rules, L, A,..., A, with increasing accuracy, see Fig. 2 for the arrangement of sample points in the right-half interval with the sample points symmetrically arranged in the left-half interval. From Fig. 2 we find that the sample points for 4 are all reused in 4+i, 0 < /c < 3. Further, they can be also reused after the bisection of the interval. See also Sugiura and Sakurai [15]. 

Fig. 2 Arrangement of sample points in the right-half interval...

For each value of P we then used a modified half-interval search to solve Equation 11 for p. For 276 experimentally measured subcritical points, the rms error in p was only 0.0567%, with systematic error occurring only at very low densities for near-critical temperatures. (Such points are omitted from the preceding rms error calculation. However, constraining a(T) to the values given by Equation 6 and the inclusion of a small residual expression in Equation 4 extend the validity of Equation 1 to extremely low densities. The fact that such accurate results were achieved using only fitted data reflects very favorably on the general accuracy of Goodwin s data tabulation.)... 

Secant. The method uses a linear approximation of the Jacobian. It may be implemented with some enhancements, as half interval option. It is recommended for single variable, discontinuous or flat convergence functions. 

When the values Lj, L2 are sufficiently large and/or the values and Df, are sufficiently small, the parameters , 2 can take any small values. It can happen that the values of and 2 re not coupled and, generally speaking, take different values. Depending on the problem formulation, we assume that the values of Ej, 2 are independent of each other and take arbitrary values in the half-interval (0,1]. 

The bisection or half-interval method is a systematic trial-and-error solution consisting of the following steps ... 

The averager operator shifts its operand by a half interval to the right of the pivot and by a half interval to the left of the pivot, evaluates it at these two positions, and averages the two values. 

Other forms of Stirling s interpolation formula exist, which make use of base points spaced at half intervals (i.e., at h/2). Our choice of using averaged central differences to replace the odd differential operators eliminated the need for having base points located at the midpoints. The central differences for Eq. (3.129) are tabulated in Table 3.6. 

The restriction is that/(a) and/(h) must have opposite signs—one of than must be positive, the other negative (it does not matter which). Then, because/is assumed to be continuous, it must be a zero somewhere in [a, b. Let c be the midpoint of [a, b]. Either c is the root, or the root lies in [a, c] or in [a, b. Ify(c) is close enough to zero (see below regarding tolerance), then the root has been found. Otherwise, one pair of /(a),/(c)]or lfic)Jlb)] has opposite signs. Keep the half-interval with opposite signs and discard the other. Repeat the process until either (1) /, evaluated at the midpoint of the interval, is sufficiently small or (2) the interval has been shrunk to a suitably small value. 

Our next task is thus to set up the microintegral equations corresponding with expression (4.55) for the u and b boundary half-intervals. We commence at Eq. (4.22) and proceed as before. In the flame context, it is assumed that no chemical reaction takes place in the volume of either of the end half-intervals, that is, that the space integral rates there are zero. This is reasonable, since there should be no major reaction near free boundaries, while walls are presumed to be perfect sinks for free radicals. 

---