BENCHMARKING AND IMPLEMENTATION OF A NEW INDEPENDENT MONTE CARLO DOSE CALCULATION QUALITY ASSURANCE AUDIT TOOL FOR CLINICAL TRIALS

Introduction Commercial treatment planning systems employ a variety of dose calculation algorithms to plan and predict the dose distributions a patient receives during external beam radiation therapy. Traditionally, the Radiological Physics Center has relied on measurements to assure that institutions participating in the National Cancer Institute sponsored clinical trials administer radiation in doses that are clinically comparable to those of other participating institutions. To complement the effort of the RPC, an independent dose calculation tool needs to be developed that will enable a generic method to determine patient dose distributions in three dimensions and to perform retrospective analysis of radiation delivered to patients who enrolled in past clinical trials. Methods A multi-source model representing output for Varian 6 MV and 10 MV photon beams was developed and evaluated. The Monte Carlo algorithm, know as the Dose Planning Method (DPM), was used to perform the dose calculations. The dose calculations were compared to measurements made in a water phantom and in anthropomorphic phantoms. Intensity modulated radiation therapy and stereotactic body radiation therapy techniques were used with the anthropomorphic phantoms. Finally, past patient treatment plans were selected and recalculated using DPM and contrasted against a commercial dose calculation algorithm. Results The multi-source model was validated for the Varian 6 MV and 10 MV photon beams. The benchmark evaluations demonstrated the ability of the model to accurately calculate dose for the Varian 6 MV and the Varian 10 MV source models. The patient calculations proved that the model was reproducible in determining dose under similar conditions described by the benchmark tests. Conclusions The dose calculation tool that relied on a multi-source model approach and used the DPM code to calculate dose was developed, validated, and benchmarked for the Varian 6 MV and 10 MV photon beams. Several patient dose distributions were contrasted against a commercial algorithm to provide a proof of principal to use as an application in monitoring clinical trial activity.

A critical evaluation of secondary cancer risk models applied to Monte Carlo dose distributions of 2-dimensional, 3-dimensional conformal and hybrid intensity-modulated radiation therapy for breast cancer

The comparison of radiotherapy techniques regarding secondary cancer risk has yielded contradictory results possibly stemming from the many different approaches used to estimate risk. The purpose of this study was to make a comprehensive evaluation of different available risk models applied to detailed whole-body dose distributions computed by Monte Carlo for various breast radiotherapy techniques including conventional open tangents, 3D conformal wedged tangents and hybrid intensity modulated radiation therapy (IMRT). First, organ-specific linear risk models developed by the International Commission on Radiological Protection (ICRP) and the Biological Effects of Ionizing Radiation (BEIR) VII committee were applied to mean doses for remote organs only and all solid organs. Then, different general non-linear risk models were applied to the whole body dose distribution. Finally, organ-specific non-linear risk models for the lung and breast were used to assess the secondary cancer risk for these two specific organs. A total of 32 different calculated absolute risks resulted in a broad range of values (between 0.1% and 48.5%) underlying the large uncertainties in absolute risk calculation. The ratio of risk between two techniques has often been proposed as a more robust assessment of risk than the absolute risk. We found that the ratio of risk between two techniques could also vary substantially considering the different approaches to risk estimation. Sometimes the ratio of risk between two techniques would range between values smaller and larger than one, which then translates into inconsistent results on the potential higher risk of one technique compared to another. We found however that the hybrid IMRT technique resulted in a systematic reduction of risk compared to the other techniques investigated even though the magnitude of this reduction varied substantially with the different approaches investigated. Based on the epidemiological data available, a reasonable approach to risk estimation would be to use organ-specific non-linear risk models applied to the dose distributions of organs within or near the treatment fields (lungs and contralateral breast in the case of breast radiotherapy) as the majority of radiation-induced secondary cancers are found in the beam-bordering regions.

Combining Monte Carlo methods with coherent wave optics for the simulation of phase-sensitive X-ray imaging

Phase-sensitive X-ray imaging shows a high sensitivity towards electron density variations, making it well suited for imaging of soft tissue matter. However, there are still open questions about the details of the image formation process. Here, a framework for numerical simulations of phase-sensitive X-ray imaging is presented, which takes both particle- and wave-like properties of X-rays into consideration. A split approach is presented where we combine a Monte Carlo method (MC) based sample part with a wave optics simulation based propagation part, leading to a framework that takes both particle- and wave-like properties into account. The framework can be adapted to different phase-sensitive imaging methods and has been validated through comparisons with experiments for grating interferometry and propagation-based imaging. The validation of the framework shows that the combination of wave optics and MC has been successfully implemented and yields good agreement between measurements and simulations. This demonstrates that the physical processes relevant for developing a deeper understanding of scattering in the context of phase-sensitive imaging are modelled in a sufficiently accurate manner. The framework can be used for the simulation of phase-sensitive X-ray imaging, for instance for the simulation of grating interferometry or propagation-based imaging.

Forward treatment planning for modulated electron radiotherapy (MERT) employing Monte Carlo methods

PURPOSE This paper describes the development of a forward planning process for modulated electron radiotherapy (MERT). The approach is based on a previously developed electron beam model used to calculate dose distributions of electron beams shaped by a photon multi leaf collimator (pMLC). METHODS As the electron beam model has already been implemented into the Swiss Monte Carlo Plan environment, the Eclipse treatment planning system (Varian Medical Systems, Palo Alto, CA) can be included in the planning process for MERT. In a first step, CT data are imported into Eclipse and a pMLC shaped electron beam is set up. This initial electron beam is then divided into segments, with the electron energy in each segment chosen according to the distal depth of the planning target volume (PTV) in beam direction. In order to improve the homogeneity of the dose distribution in the PTV, a feathering process (Gaussian edge feathering) is launched, which results in a number of feathered segments. For each of these segments a dose calculation is performed employing the in-house developed electron beam model along with the macro Monte Carlo dose calculation algorithm. Finally, an automated weight optimization of all segments is carried out and the total dose distribution is read back into Eclipse for display and evaluation. One academic and two clinical situations are investigated for possible benefits of MERT treatment compared to standard treatments performed in our clinics and treatment with a bolus electron conformal (BolusECT) method. RESULTS The MERT treatment plan of the academic case was superior to the standard single segment electron treatment plan in terms of organs at risk (OAR) sparing. Further, a comparison between an unfeathered and a feathered MERT plan showed better PTV coverage and homogeneity for the feathered plan, with V95% increased from 90% to 96% and V107% decreased from 8% to nearly 0%. For a clinical breast boost irradiation, the MERT plan led to a similar homogeneity in the PTV compared to the standard treatment plan while the mean body dose was lower for the MERT plan. Regarding the second clinical case, a whole breast treatment, MERT resulted in a reduction of the lung volume receiving more than 45% of the prescribed dose when compared to the standard plan. On the other hand, the MERT plan leads to a larger low-dose lung volume and a degraded dose homogeneity in the PTV. For the clinical cases evaluated in this work, treatment plans using the BolusECT technique resulted in a more homogenous PTV and CTV coverage but higher doses to the OARs than the MERT plans. CONCLUSIONS MERT treatments were successfully planned for phantom and clinical cases, applying a newly developed intuitive and efficient forward planning strategy that employs a MC based electron beam model for pMLC shaped electron beams. It is shown that MERT can lead to a dose reduction in OARs compared to other methods. The process of feathering MERT segments results in an improvement of the dose homogeneity in the PTV.

A stochastic model updating method for parameter variability quantification based on response surface models and Monte Carlo simulation

Stochastic model updating must be considered for quantifying uncertainties inherently existing in real-world engineering structures. By this means the statistical properties,instead of deterministic values, of structural parameters can be sought indicating the parameter variability. However, the implementation of stochastic model updating is much more complicated than that of deterministic methods particularly in the aspects of theoretical complexity and low computational efficiency. This study attempts to propose a simple and cost-efficient method by decomposing a stochastic updating process into a series of deterministic ones with the aid of response surface models and Monte Carlo simulation. The response surface models are used as surrogates for original FE models in the interest of programming simplification, fast response computation and easy inverse optimization. Monte Carlo simulation is adopted for generating samples from the assumed or measured probability distributions of responses. Each sample corresponds to an individual deterministic inverse process predicting the deterministic values of parameters. Then the parameter means and variances can be statistically estimated based on all the parameter predictions by running all the samples. Meanwhile, the analysis of variance approach is employed for the evaluation of parameter variability significance. The proposed method has been demonstrated firstly on a numerical beam and then a set of nominally identical steel plates tested in the laboratory. It is found that compared with the existing stochastic model updating methods, the proposed method presents similar accuracy while its primary merits consist in its simple implementation and cost efficiency in response computation and inverse optimization.

Monte Carlo uncertainty propagation approaches in ADS burn-up calculations

In activation calculations, there are several approaches to quantify uncertainties: deterministic by means of sensitivity analysis, and stochastic by means of Monte Carlo. Here, two different Monte Carlo approaches for nuclear data uncertainty are presented: the first one is the Total Monte Carlo (TMC). The second one is by means of a Monte Carlo sampling of the covariance information included in the nuclear data libraries to propagate these uncertainties throughout the activation calculations. This last approach is what we named Covariance Uncertainty Propagation, CUP. This work presents both approaches and their differences. Also, they are compared by means of an activation calculation, where the cross-section uncertainties of 239Pu and 241Pu are propagated in an ADS activation calculation.

Massively parallelizable list-mode reconstruction using a Monte Carlo-based elliptical Gaussian model

Purpose: A fully three-dimensional (3D) massively parallelizable list-mode ordered-subsets expectation-maximization (LM-OSEM) reconstruction algorithm has been developed for high-resolution PET cameras. System response probabilities are calculated online from a set of parameters derived from Monte Carlo simulations. The shape of a system response for a given line of response (LOR) has been shown to be asymmetrical around the LOR. This work has been focused on the development of efficient region-search techniques to sample the system response probabilities, which are suitable for asymmetric kernel models, including elliptical Gaussian models that allow for high accuracy and high parallelization efficiency. The novel region-search scheme using variable kernel models is applied in the proposed PET reconstruction algorithm. Methods: A novel region-search technique has been used to sample the probability density function in correspondence with a small dynamic subset of the field of view that constitutes the region of response (ROR). The ROR is identified around the LOR by searching for any voxel within a dynamically calculated contour. The contour condition is currently defined as a fixed threshold over the posterior probability, and arbitrary kernel models can be applied using a numerical approach. The processing of the LORs is distributed in batches among the available computing devices, then, individual LORs are processed within different processing units. In this way, both multicore and multiple many-core processing units can be efficiently exploited. Tests have been conducted with probability models that take into account the noncolinearity, positron range, and crystal penetration effects, that produced tubes of response with varying elliptical sections whose axes were a function of the crystal's thickness and angle of incidence of the given LOR. The algorithm treats the probability model as a 3D scalar field defined within a reference system aligned with the ideal LOR. Results: This new technique provides superior image quality in terms of signal-to-noise ratio as compared with the histogram-mode method based on precomputed system matrices available for a commercial small animal scanner. Reconstruction times can be kept low with the use of multicore, many-core architectures, including multiple graphic processing units. Conclusions: A highly parallelizable LM reconstruction method has been proposed based on Monte Carlo simulations and new parallelization techniques aimed at improving the reconstruction speed and the image signal-to-noise of a given OSEM algorithm. The method has been validated using simulated and real phantoms. A special advantage of the new method is the possibility of defining dynamically the cut-off threshold over the calculated probabilities thus allowing for a direct control on the trade-off between speed and quality during the reconstruction.

Monte Carlo Neutronics and Thermal-hydraulics coupling applied to Fast Reactors

Un escenario habitualmente considerado para el uso sostenible y prolongado de la energía nuclear contempla un parque de reactores rápidos refrigerados por metales líquidos (LMFR) dedicados al reciclado de Pu y la transmutación de actínidos minoritarios (MA). Otra opción es combinar dichos reactores con algunos sistemas subcríticos asistidos por acelerador (ADS), exclusivamente destinados a la eliminación de MA. El diseño y licenciamiento de estos reactores innovadores requiere herramientas computacionales prácticas y precisas, que incorporen el conocimiento obtenido en la investigación experimental de nuevas configuraciones de reactores, materiales y sistemas. A pesar de que se han construido y operado un cierto número de reactores rápidos a nivel mundial, la experiencia operacional es todavía reducida y no todos los transitorios se han podido entender completamente. Por tanto, los análisis de seguridad de nuevos LMFR están basados fundamentalmente en métodos deterministas, al contrario que las aproximaciones modernas para reactores de agua ligera (LWR), que se benefician también de los métodos probabilistas. La aproximación más usada en los estudios de seguridad de LMFR es utilizar una variedad de códigos, desarrollados a base de distintas teorías, en busca de soluciones integrales para los transitorios e incluyendo incertidumbres. En este marco, los nuevos códigos para cálculos de mejor estimación ("best estimate") que no incluyen aproximaciones conservadoras, son de una importancia primordial para analizar estacionarios y transitorios en reactores rápidos. Esta tesis se centra en el desarrollo de un código acoplado para realizar análisis realistas en reactores rápidos críticos aplicando el método de Monte Carlo. Hoy en día, dado el mayor potencial de recursos computacionales, los códigos de transporte neutrónico por Monte Carlo se pueden usar de manera práctica para realizar cálculos detallados de núcleos completos, incluso de elevada heterogeneidad material. Además, los códigos de Monte Carlo se toman normalmente como referencia para los códigos deterministas de difusión en multigrupos en aplicaciones con reactores rápidos, porque usan secciones eficaces punto a punto, un modelo geométrico exacto y tienen en cuenta intrínsecamente la dependencia angular de flujo. En esta tesis se presenta una metodología de acoplamiento entre el conocido código MCNP, que calcula la generación de potencia en el reactor, y el código de termohidráulica de subcanal COBRA-IV, que obtiene las distribuciones de temperatura y densidad en el sistema. COBRA-IV es un código apropiado para aplicaciones en reactores rápidos ya que ha sido validado con resultados experimentales en haces de barras con sodio, incluyendo las correlaciones más apropiadas para metales líquidos. En una primera fase de la tesis, ambos códigos se han acoplado en estado estacionario utilizando un método iterativo con intercambio de archivos externos. El principal problema en el acoplamiento neutrónico y termohidráulico en estacionario con códigos de Monte Carlo es la manipulación de las secciones eficaces para tener en cuenta el ensanchamiento Doppler cuando la temperatura del combustible aumenta. Entre todas las opciones disponibles, en esta tesis se ha escogido la aproximación de pseudo materiales, y se ha comprobado que proporciona resultados aceptables en su aplicación con reactores rápidos. Por otro lado, los cambios geométricos originados por grandes gradientes de temperatura en el núcleo de reactores rápidos resultan importantes para la neutrónica como consecuencia del elevado recorrido libre medio del neutrón en estos sistemas. Por tanto, se ha desarrollado un módulo adicional que simula la geometría del reactor en caliente y permite estimar la reactividad debido a la expansión del núcleo en un transitorio. éste módulo calcula automáticamente la longitud del combustible, el radio de la vaina, la separación de los elementos de combustible y el radio de la placa soporte en función de la temperatura. éste efecto es muy relevante en transitorios sin inserción de bancos de parada. También relacionado con los cambios geométricos, se ha implementado una herramienta que, automatiza el movimiento de las barras de control en busca d la criticidad del reactor, o bien calcula el valor de inserción axial las barras de control. Una segunda fase en la plataforma de cálculo que se ha desarrollado es la simulació dinámica. Puesto que MCNP sólo realiza cálculos estacionarios para sistemas críticos o supercríticos, la solución más directa que se propone sin modificar el código fuente de MCNP es usar la aproximación de factorización de flujo, que resuelve por separado la forma del flujo y la amplitud. En este caso se han estudiado en profundidad dos aproximaciones: adiabática y quasiestática. El método adiabático usa un esquema de acoplamiento que alterna en el tiempo los cálculos neutrónicos y termohidráulicos. MCNP calcula el modo fundamental de la distribución de neutrones y la reactividad al final de cada paso de tiempo, y COBRA-IV calcula las propiedades térmicas en el punto intermedio de los pasos de tiempo. La evolución de la amplitud de flujo se calcula resolviendo las ecuaciones de cinética puntual. Este método calcula la reactividad estática en cada paso de tiempo que, en general, difiere de la reactividad dinámica que se obtendría con la distribución de flujo exacta y dependiente de tiempo. No obstante, para entornos no excesivamente alejados de la criticidad ambas reactividades son similares y el método conduce a resultados prácticos aceptables. Siguiendo esta línea, se ha desarrollado después un método mejorado para intentar tener en cuenta el efecto de la fuente de neutrones retardados en la evolución de la forma del flujo durante el transitorio. El esquema consiste en realizar un cálculo cuasiestacionario por cada paso de tiempo con MCNP. La simulación cuasiestacionaria se basa EN la aproximación de fuente constante de neutrones retardados, y consiste en dar un determinado peso o importancia a cada ciclo computacial del cálculo de criticidad con MCNP para la estimación del flujo final. Ambos métodos se han verificado tomando como referencia los resultados del código de difusión COBAYA3 frente a un ejercicio común y suficientemente significativo. Finalmente, con objeto de demostrar la posibilidad de uso práctico del código, se ha simulado un transitorio en el concepto de reactor crítico en fase de diseño MYRRHA/FASTEF, de 100 MW de potencia térmica y refrigerado por plomo-bismuto. ABSTRACT Long term sustainable nuclear energy scenarios envisage a fleet of Liquid Metal Fast Reactors (LMFR) for the Pu recycling and minor actinides (MAs) transmutation or combined with some accelerator driven systems (ADS) just for MAs elimination. Design and licensing of these innovative reactor concepts require accurate computational tools, implementing the knowledge obtained in experimental research for new reactor configurations, materials and associated systems. Although a number of fast reactor systems have already been built, the operational experience is still reduced, especially for lead reactors, and not all the transients are fully understood. The safety analysis approach for LMFR is therefore based only on deterministic methods, different from modern approach for Light Water Reactors (LWR) which also benefit from probabilistic methods. Usually, the approach adopted in LMFR safety assessments is to employ a variety of codes, somewhat different for the each other, to analyze transients looking for a comprehensive solution and including uncertainties. In this frame, new best estimate simulation codes are of prime importance in order to analyze fast reactors steady state and transients. This thesis is focused on the development of a coupled code system for best estimate analysis in fast critical reactor. Currently due to the increase in the computational resources, Monte Carlo methods for neutrons transport can be used for detailed full core calculations. Furthermore, Monte Carlo codes are usually taken as reference for deterministic diffusion multigroups codes in fast reactors applications because they employ point-wise cross sections in an exact geometry model and intrinsically account for directional dependence of the ux. The coupling methodology presented here uses MCNP to calculate the power deposition within the reactor. The subchannel code COBRA-IV calculates the temperature and density distribution within the reactor. COBRA-IV is suitable for fast reactors applications because it has been validated against experimental results in sodium rod bundles. The proper correlations for liquid metal applications have been added to the thermal-hydraulics program. Both codes are coupled at steady state using an iterative method and external files exchange. The main issue in the Monte Carlo/thermal-hydraulics steady state coupling is the cross section handling to take into account Doppler broadening when temperature rises. Among every available options, the pseudo materials approach has been chosen in this thesis. This approach obtains reasonable results in fast reactor applications. Furthermore, geometrical changes caused by large temperature gradients in the core, are of major importance in fast reactor due to the large neutron mean free path. An additional module has therefore been included in order to simulate the reactor geometry in hot state or to estimate the reactivity due to core expansion in a transient. The module automatically calculates the fuel length, cladding radius, fuel assembly pitch and diagrid radius with the temperature. This effect will be crucial in some unprotected transients. Also related to geometrical changes, an automatic control rod movement feature has been implemented in order to achieve a just critical reactor or to calculate control rod worth. A step forward in the coupling platform is the dynamic simulation. Since MCNP performs only steady state calculations for critical systems, the more straight forward option without modifying MCNP source code, is to use the flux factorization approach solving separately the flux shape and amplitude. In this thesis two options have been studied to tackle time dependent neutronic simulations using a Monte Carlo code: adiabatic and quasistatic methods. The adiabatic methods uses a staggered time coupling scheme for the time advance of neutronics and the thermal-hydraulics calculations. MCNP computes the fundamental mode of the neutron flux distribution and the reactivity at the end of each time step and COBRA-IV the thermal properties at half of the the time steps. To calculate the flux amplitude evolution a solver of the point kinetics equations is used. This method calculates the static reactivity in each time step that in general is different from the dynamic reactivity calculated with the exact flux distribution. Nevertheless, for close to critical situations, both reactivities are similar and the method leads to acceptable practical results. In this line, an improved method as an attempt to take into account the effect of delayed neutron source in the transient flux shape evolutions is developed. The scheme performs a quasistationary calculation per time step with MCNP. This quasistationary simulations is based con the constant delayed source approach, taking into account the importance of each criticality cycle in the final flux estimation. Both adiabatic and quasistatic methods have been verified against the diffusion code COBAYA3, using a theoretical kinetic exercise. Finally, a transient in a critical 100 MWth lead-bismuth-eutectic reactor concept is analyzed using the adiabatic method as an application example in a real system.

Efficient Monte Carlo optimization for multi-label classifier chains

Multi-label classification (MLC) is the supervised learning problem where an instance may be associated with multiple labels. Modeling dependencies between labels allows MLC methods to improve their performance at the expense of an increased computational cost. In this paper we focus on the classifier chains (CC) approach for modeling dependencies. On the one hand, the original CC algorithm makes a greedy approximation, and is fast but tends to propagate errors down the chain. On the other hand, a recent Bayes-optimal method improves the performance, but is computationally intractable in practice. Here we present a novel double-Monte Carlo scheme (M2CC), both for finding a good chain sequence and performing efficient inference. The M2CC algorithm remains tractable for high-dimensional data sets and obtains the best overall accuracy, as shown on several real data sets with input dimension as high as 1449 and up to 103 labels.

Efficient Monte Carlo methods for multi-dimensional learning with classifier chains

Multi-dimensional classification (MDC) is the supervised learning problem where an instance is associated with multiple classes, rather than with a single class, as in traditional classification problems. Since these classes are often strongly correlated, modeling the dependencies between them allows MDC methods to improve their performance – at the expense of an increased computational cost. In this paper we focus on the classifier chains (CC) approach for modeling dependencies, one of the most popular and highest-performing methods for multi-label classification (MLC), a particular case of MDC which involves only binary classes (i.e., labels). The original CC algorithm makes a greedy approximation, and is fast but tends to propagate errors along the chain. Here we present novel Monte Carlo schemes, both for finding a good chain sequence and performing efficient inference. Our algorithms remain tractable for high-dimensional data sets and obtain the best predictive performance across several real data sets.

Monte Carlo limit cycle characterization

The fixed point implementation of IIR digital filters usually leads to the appearance of zero-input limit cycles, which degrade the performance of the system. In this paper, we develop an efficient Monte Carlo algorithm to detect and characterize limit cycles in fixed-point IIR digital filters. The proposed approach considers filters formulated in the state space and is valid for any fixed point representation and quantization function. Numerical simulations on several high-order filters, where an exhaustive search is unfeasible, show the effectiveness of the proposed approach.

Reconfigurable computing for Monte Carlo simulations: results and prospects of the Janus project

We describe Janus, a massively parallel FPGA-based computer optimized for the simulation of spin glasses, theoretical models for the behavior of glassy materials. FPGAs (as compared to GPUs or many-core processors) provide a complementary approach to massively parallel computing. In particular, our model problem is formulated in terms of binary variables, and floating-point operations can be (almost) completely avoided. The FPGA architecture allows us to run many independent threads with almost no latencies in memory access, thus updating up to 1024 spins per cycle. We describe Janus in detail and we summarize the physics results obtained in four years of operation of this machine; we discuss two types of physics applications: long simulations on very large systems (which try to mimic and provide understanding about the experimental non equilibrium dynamics), and low-temperature equilibrium simulations using an artificial parallel tempering dynamics. The time scale of our non-equilibrium simulations spans eleven orders of magnitude (from picoseconds to a tenth of a second). On the other hand, our equilibrium simulations are unprecedented both because of the low temperatures reached and for the large systems that we have brought to equilibrium. A finite-time scaling ansatz emerges from the detailed comparison of the two sets of simulations. Janus has made it possible to perform spin glass simulations that would take several decades on more conventional architectures. The paper ends with an assessment of the potential of possible future versions of the Janus architecture, based on state-of-the-art technology.

Searching for convergence in phylogenetic Markov chain Monte Carlo

Markov chain Monte Carlo (MCMC) is a methodology that is gaining widespread use in the phylogenetics community and is central to phylogenetic software packages such as MrBayes. An important issue for users of MCMC methods is how to select appropriate values for adjustable parameters such as the length of the Markov chain or chains, the sampling density, the proposal mechanism, and, if Metropolis-coupled MCMC is being used, the number of heated chains and their temperatures. Although some parameter settings have been examined in detail in the literature, others are frequently chosen with more regard to computational time or personal experience with other data sets. Such choices may lead to inadequate sampling of tree space or an inefficient use of computational resources. We performed a detailed study of convergence and mixing for 70 randomly selected, putatively orthologous protein sets with different sizes and taxonomic compositions. Replicated runs from multiple random starting points permit a more rigorous assessment of convergence, and we developed two novel statistics, delta and epsilon, for this purpose. Although likelihood values invariably stabilized quickly, adequate sampling of the posterior distribution of tree topologies took considerably longer. Our results suggest that multimodality is common for data sets with 30 or more taxa and that this results in slow convergence and mixing. However, we also found that the pragmatic approach of combining data from several short, replicated runs into a metachain to estimate bipartition posterior probabilities provided good approximations, and that such estimates were no worse in approximating a reference posterior distribution than those obtained using a single long run of the same length as the metachain. Precision appears to be best when heated Markov chains have low temperatures, whereas chains with high temperatures appear to sample trees with high posterior probabilities only rarely. [Bayesian phylogenetic inference; heating parameter; Markov chain Monte Carlo; replicated chains.]

Monte Carlo Study of Particle Transport Problem in Air Pollution

2002 Mathematics Subject Classification: 65C05.