Published electron microprobe analyses of mattheddleite, a lead sulpho-silicate apatite from Leadhills, Scotland, have 9-13% IV site deficiencies. However, galena was used as a standard for S, which suggested that low S resulted from a shift in the S-K alpha peak. Wavelength scans with a PET crystal show that the S-K alpha peak is shifted down by 0.0026 angstrom for sulphates relative to sulphides. Quantitative analyses show a similar to 30% increase of S in mattheddleite using a celestite standard, which fills the IV site, but with Si > S, on average Pb5S1.2Si1.8O11.7Cl0.6(OH)(0.4). Direct analysis of oxygen with the electron microprobe implies that the charge imbalance engendered from the inequality of Si and S is compensated with substitution of a vacancy (square), as in Pb5S1.2Si1.8[O-11.7 square(0.3)][Cl-0.6(OH)(0.4)] or Pb5S1.2Si1.8[O-11.7(Cl,OH)(0.3)][Cl,OH)(0.7)square(0.3)]. Calculation of OH as l-Cl suggests the presence of both OH- and Cl-dominant mattheddleite at Leadhills, but direct analysis of H is needed to confirm the dominance of OH in the channel site. Wavelength-dispersive analyses of S in apatite and other sulphates must be undertaken with sulphate standards: use of sulphide standards yields a negative error on the order of 10-20% in the resultant S concentration. Reactions of mattheddleite with other Pb minerals at Leadhills show that their stability depends on fluid composition as well as pressure and temperature. An X-ray map of Cl shows complex zoning between Cl-poor and Cl-rich mattheddleite, recording rapid changes in the fluid chemistry during late-stage hydrothermal processes at Leadhills.

Aflatoxin contamination in crops of peanuts is a problem of significant health and financial importance, so it would be useful to develop techniques to predict the levels prior to harvest. Backpropagation neural networks have been used in the past to model problems of this type, however development of networks poses the complex problem of setting values for architectural features and backpropagation parameters. Genetic algorithms have been used in prior efforts to locate parameters for backpropagation neural networks. This paper describes the development of a genetic algorithm/backpropagation neural network hybrid (GA/BPN) in which a genetic algorithm is used to find architectures and backpropagation parameter values simultaneously for a backpropagation neural network that predicts aflatoxin contamination levels in peanuts based on environmental data.

Aflatoxin contamination in peanut crops is a problem of significant health and financial importance. Predicting aflatoxin levels prior to crop harvest is useful for minimizing the impact of a contaminated crop and is the goal of our research. Backpropagation neural networks have been used to model problems of this type, however development of networks poses the complex problem of setting values for architectural features and backpropagation parameters. Genetic algorithms have been used in other studies to determine parameters for backpropagation neural networks. This paper describes the development of a genetic algorithm/backpropagation neural network hybrid (GA/BPN) in which a genetic algorithm is used to find architectures and backpropagation parameter values simultaneously for a backpropagation neural network that predicts aflatoxin contamination levels in peanuts based on environmental data. Learning rate, momentum, and number of hidden nodes are the parameters that are set by the genetic algorithm. A three-layer feed-forward network with logistic activation functions is used. Inputs to the network are soil temperature, drought duration, crop age, and accumulated heat units. The project showed that the GA/BPN approach automatically finds highly fit parameter sets for backpropagation neural networks for the aflatoxin problem.

Quantitative analysis with the electron microprobe analyzer (EMPA) has yielded more accurate results over time as a result of improvements in ZAF and other correction routines, mass absorption coefficients, synthetic pseudocrystals for ultralight elements, computers, software programs, backscattered electron (BSE) and energy dispersive (EDS) X-ray detectors. Consequently, many geoscientists view EMPA as routine, and details of procedures, standards, and operating conditions are seldom provided in current publications. However, in overseeing a facility with many users, we have learned that acceptable analytical data are sometimes difficult to obtain even with established analytical procedures and a choice of several hundred standards. After novice users have mastered the routines of sample polishing, cleaning, coating, handling and machine focus, their choice of nonoptimal standards often prevents them from obtaining the most accurate results possible. Optimal analysis for geological problems requires choosing appropriate standards, selection of optimal operating conditions, as well as consideration of the possibility of omitted elements, peak and background overlaps, matrix absorption effects, beam damage and elemental migration, reintegration of heterogeneous materials, fluorescence effects, and variations in the oxidation state of iron.

Predicting the level of aflatoxin contamination in crops of peanuts is a task of significant importance. Backpmpagation neural networks have been used in the past to model this problem, but use of the backpropagation algorithm for training introduces limitations and difficulties. Therefore, it is useful to explore alternative learning algorithms. Genetic algorithms provide an effective technique for searching large spaces, and have been used in the past to train neural networks. This paper describes the development of a genetic algorithm/neural network hybrid in which a genetic algorithm is used to find weight assignments for a neural network that predicts aflatoxin contamination levels in peanuts based on environmental data. Genetic algorithms are a method for searching large, complex spaces, and have been used to train neural networks. By linking the global search capabilities of the genetic algorithm with the modeling power of artificial neural networks, a highly effective predietive tool for the aflatoxin problem can be constructed. This paper describes the use of a genetic algorithm/neural network hybrid that trains networks for the aflatoxin problem. Fixed architecture, three layer feedforward networks with logistic activation functions were used to develop the models. As in a previous study by Parmar et al. (1997), inputs for the networks were soil temperature, drought duration, crop age, and accumulated heat units. The value for accumulated heat units was calculated based on a threshold temperature of 25°c. There were two neural network models developed for the problem. One network (model A) was trained on all available data, while the other (model B) was trained using only data from undamaged peanuts. The network for model A produced rz values of 0.74, 0.83, and 0.21 for training, testing, and validation data sets. The model B network produced values of 0.45, 0.82, and 0.41. The implementation demonstrated that genetic algorithms can be used to train effective networks for the aflatoxin problem. The genetic search has the advantage that setting parameters is easier than for baekpropagation, and the genetic trainer does not suffer from many backpropagation shortcomings, such as the tendency to stick in local minima.

Adult drones were removed from experimental colonies and the number of drones reared was compared with that of control colonies containing natural numbers of drones. No differences were found in the number of drones reared in the experimental and control colonies in 3 samplings taken at 12, 24 and 36 d post-treatment.

Forty honey bee eggs from two unrelated, similar-sized colonies were measured weekly from 1 June to 12 August 1990, in Ithaca, NY, USA. The eggs were obtained from a section of comb, containing both worker- and drone-sized cells, on which the queen was confined for 24 h. Egg length and width fluctuated significantly in both colonies throughout the experimental period. There was no statistical difference between sexes, and eggs destined to produce drones during the swarming season were not necessarily larger.Dry weights of drones were measured at 14-day intervals during spring (1990, Lake Placid, FL) and summer (1989 and 1990, Ithaca, NY), using a pool of 12-15 colonies. Weights varied significantly throughout all the sampling periods. The 1989 (NY) and 1990 (FL) data showed a tendency for drones to become lighter after swarming and heavier during swarming. The more extensive sampling in 1990 (NY) showed a mean drone dry weight varying from 56.86 mg to 30.71 mg, but there was no association with swarming and similar changes occurred also in worker weights.