quired. A mean square error (MSE) estimate of 1.3 was obtained for GB; 0.4 for the pyroester. B y correcting for the average loss of GB in the column, the MSE can be reduced to 0.6. Data were obtained using purified G B with the best available sample of the diester pyrophosphonate as simulated mixtures. Where crude GB containing the pyroester was used, one test was made maintaining the relative concentrations of the two compounds. The second test was made keeping the GB as a constant quantity while varying the pyroester. The results are shown in Table VI. Comparative results obtained using the silica gel column method and the less accurate, but more rapid, liquid-liquid method are shown in Table VII. The rapid or free acid titration to the first methyl red end point can be used to estimate the total of phosphorus and halide acids present as impurity. However, as true acidity, this estimate is complicated by the possible presence of the readily hydrolyzable phosphorus dihalidates. These latter compounds would also be titrated as acid in the aqueous system. Precision and Accuracy. A calculated standard deviation of 0.27 was obtained for t h e hydrolytic method described in this report. This result is based on d a t a obtained by three research analytical chemists using a t least three different samples of most of t h e mentioned compounds in replicates of five to ten determinations. I n the case of GB, the data represent more than 100 different samples (some varying in purity by 40%), more chemists, but less replication per sample. When tested on the basis of the individual experience of technicians with the method, the standard deviation de-
Table VII.
Comparison of Silica Gel Column and liquid-Liquid Partitioning Methods for Quantitative Determination of Pyrodiester in GBQ
Pyrodiester Added to GB Sample, yo by Weight
Pyrodiester Found, yo by Weight Silica Gel Method
Recovery,
%
Pyrodiester Found, by Weight Liquid-Liquid Partitioning Method
