4. METHOD OF SOLUTION
As input data the program needs the probability distributions of the variable parameters (e.g., reactivity and heat transfer coefficients considered as sources of uncertainties in an accident analysis). Uniform, normal, truncated normal, exponential, beta, and log-normal probability distributions are available.
The first part of the program determines knot-point coordinates of the parameters, using a user-specified probability range, or confidence interval. These serve as input to deterministic codes, such as accident analysis codes, SACO or SAS, which calculate consequence values in the specified knot points.
The consequence values are used by the second part of PROSA1 to solve 1) the approximating functions or response surfaces; 2) the sensitivity/importance of each parameter with respect to the consequence variables; 3) the statistical moments of the input parameters; 4) the mean values and standard deviations of the consequences; and 5) the correlation coefficients for all pairs of the consequences. In the simplest case a single multivariate second- degree surface is matched to knot-point consequences.
It is possible to use regionwise response surfaces, distinct second-degree surfaces for every "quadrant" of the multivariate parameter space.
By use of the random-number sampling of the parameters distributions and simulation with the as-calculated response surface the program also calculates 6) the probability distributions and the first four moments of the consequences; 7) the joint distribution; and 8) the statistical-error estimates for the distributions.