No abstract

[1] The Hawaii Scientific Drilling Program (HSDP) cored and recovered igneous rock from the surface to a depth of 3109 m near Hilo, Hawaii. Much of the deeper parts of the hole is composed of hyaloclastite (fractured basalt glass that has been cemented in situ with secondary minerals). Some hyaloclastite units have been altered in a manner attributed to microorganisms in volcanic rocks. Samples from one such unit (1336 m to 1404 m below sea level) were examined to test the hypothesis that the alteration was associated with microorganisms. Deep ultraviolet native fluorescence and resonance Raman spectroscopy indicate that nucleic acids and aromatic amino acids are present in clay inside spherical cavities (vesicles) within basalt glass. Chemical mapping shows that phosphorus and carbon were enriched at the boundary between the clay and volcanic glass of the vesicles. Environmental scanning electron microscopy (ESEM) reveals two to three micrometer coccoid structures in these same boundaries. ESEM-linked energy dispersive spectroscopy demonstrated carbon, phosphorous, chloride, and magnesium in these bodies significantly differing from unoccupied neighboring regions of basalt. These observations taken together indicate the presence of microorganisms at the boundary between primary volcanic glass and secondary clays. Amino acids and nucleic acids were extracted from bulk samples of the hyaloclastite unit. Amino acid abundance was low, and if the amino acids are derived from microorganisms in the rock, then there are less than 100,000 cells per gram of rock. Most nucleic acid sequences extracted from the unit were closely related to sequences of Crenarchaeota collected from the subsurface of the ocean floor.

Quantitative morphological classication of galaxies is important for understanding the origin of type frequency and correlations with environment. But galaxy morphological classication is still mainly done visually by dedicated individuals, in the spirit of Hubble's original scheme, and its modications. The rapid increase in data on galaxy images 1 at low and high redshift calls for re-examination of the classication schemes and for new automatic methods. Here we show results from the rst systematic comparison of the dispersion among human experts classifying a uniformly selected sample of over 800 digitised galaxy images. These galaxy images were then classied by six of the authors independently. The human classications are compared with each other, and with an automatic classication by Articial Neural Networks (ANN). It is shown that the ANNs can replicate the classication by a h uman expert to the same degree of agreement as that between two h uman experts.Hubble (1) suggested a classication scheme for galaxies which consists of a sequence starting from elliptical galaxies (E) , through lenticular (S0), to spiral galaxies (S), and a parallel branch of spirals with a barred component, leading to the so called`tuning fork' Hubble diagram. This scheme has been extended by astronomers over the years (2{5), to incorporate features such as the strength of the spiral arms, yielding multi-dimensional classication (3; 5). It is remarkable that these somewhat subjective classication labels for galaxies (as seen projected on the sky) correlate well with physical properties such a s colour, dynamical properties (e.g. rotation curves and stellar velocity dispersions) and the mass in neutral hydrogen (6). However, one would like e v entually to devise a scheme of classication, which can be related to the physical processes of galaxy formation. While there have been in recent y ears signicant advances in observational techniques (e.g. telescopes, detectors and reduction algorithms) as well as in theoretical modelling (e.g. N-body and hydrodynamics simulations), galaxy classication remains a subjective area.Quantifying galaxy morphology is important for various reasons. First, it provides important clues to the origin of galaxies and their formation processes. For example, ellipticals and lenticular galaxies comprise only 20% of the galaxies, and there is a striking density-morphology relation (1; 7), indicating that elliptical galaxies mainly re-2 side in high-density regions. Understanding the origin of the type frequency and the density-morphology relation is clearly of fundamental importance. But quantifying these properties requires reliable classication schemes. Second, galaxies can also be used e.g. to measure redshift-independent distances by methods such as the luminosity-rotation velocity relation for spirals (8) and the diameter-velocity dispersion for ellipticals (9). Clearly any observational programme requires an a priori target list of objects for photometric or spectrographic measurements. Therefore galaxy cl...

We apply and compare various Arti cial Neural Network (ANN) and other algorithms for automatic morphological classi cation of galaxies. The ANNs are presented here mathematically, as non-linear extensions of conventional statistical methods in Astronomy. The methods are illustrated using di erent subsets from the ESO-LV catalogue, for which both machine parameters and human classi cation are available. The main methods we explore are: (i) Principal Component Analysis (PCA) which tells how independent and informative the input parameters are. (ii) Encoder Neural Network which allows us to nd both linear (PCA-like) and non-linear combinations of the input, illustrating an example of unsupervised ANN. (iii) Supervised ANN (using the Backpropagation or Quasi-Newton algorithms) based on a training set for which the human classi cation is known. Here the output for previously unclassi ed galaxies can be interpreted as either a continuous (analog) output (e.g. T-type) or a Bayesian a posteriori probability for each class. Although the ESO-LV parameters are sub-optimal, the success of the ANN in reproducing the human classi cation is 2 T-type units, similar to the degree of agreement between two human experts who classify the same galaxy images on plate material. We also examine the aspects of ANN con gurations, reproducibility, scaling of input parameters and redshift information.