We are happy to release of version 2.0 of the pymbar scripts, a
significantly improved implementation of the multistate Bennett's
acceptance ratio formulation for computing free energy differences and
ensemble averages using data sampled from an arbitrary number of
thermodynamic ensembles.

Version 2.0 features:

* Greatly increased performance, at least an order of magnitude, and in
  some cases up to three orders of magnitude faster compared to the
  previous release.

  The speedup is due to a combination of a number of many different
  improvements, including:
  * optimized order of operations in Python
  * reduced recalculation of expensive quantities
  * increased use of matrix algebra 
  * introduction of the adaptive minimizer for the likelihood function.
  * initialization of MBAR free energies from pairwise BAR free energies.

  Running the examples/harmonic-oscillators.py test included in
  pymbar-1.0d, the previous version takes 46 seconds using
  Newton-Raphson algorithm, and 5528 seconds using the more robust
  self-consistent iteration.  The new code, using a robust adaptive
  minimizer that intelligently switches between self-consistent
  iteration and Newton-Raphson minimization, takes 5 seconds.  All
  cases uses the same C++ accelerated inner loop and are run on a 2011
  MacBook Pro laptop.

  If we increase the number of samples per state in the example test,
  from 40 to 4000, the new code takes 107 seconds, while the old code
  takes 4420 seconds, even using Newton-Raphson.

* Improved robustness:

  * A new adaptive minimizer is now the default, switching from
    self-consistent iteration to Newton-Raphson minimization when
    approaching the minimum.
  * All intermediate data is stored in logarithmic form and
    calculations are performed as sums of logarithms rather than
    products whenever possible, including averages.  This results in
    many fewer cases of over- and underflow.

* Improved testing and validation:
  
  * All functions now include doctests. 
  * Comparison to analytical results (including tests of the error
    estimate) are included for all functions exposed for external
    calls in tests/harmonic_oscillators/harmonic-oscillators.py.  Test
    for the timeseries functions are provided in
    tests/timeseries/test-statistical-inefficiency.py. Tests for the
    validity of the variance estimate are provided in
    tests/harmonic_oscillators/harmonic-oscillators-distributions.py

* Additional functionality:
  
  * computeOverlap: computes a measure of how much the K states overlap to
    aid in evaluating whether state are sufficiently close.
  * computeMultipleExpectations: Simplifies computation averages of 
    multiple observables at a single thermodynamic state from data 
    collected at all states, along with covariances between these observables.
  * computePerturbedExpectation: Computes expectations at a single new state 
    using samples previously collected.
  * Allows more data (such as the full covariance matrix) to be returned from all 
    expectation routines.

* Improved examples:

  * An example for generating the single molecule
    constant-force-optical-trap in the original Shirts and Chodera
    2008 JCP MBAR paper is included.

  * alchemical-gromacs.py is updated for Gromacs 4.6 output, and
    computes a number of different free energy estimators for
    comparison to MBAR.

  * heat-capacity.py for computing heat capacity from energies
    collected at multiple temperatures

* Automated setup

  * C++ accelerated code is now automatically accelerated and pymbar
    is added to site-packages

Changes between version 2.0beta and version 2.0:

  * Allowing more data (such as the full covariance matrix) to be returned from all
    expectation	routines, making them more consistent.
  * adding the heat capacity example.
  * fixing bugs in some doctests
  * fixing data type in return values for expectation functions.
  * alchemical-gromacs.py example and example data adjusted for latest output file formats in Gromacs 4.6.
  * Time series import warning only imported once if timeseries code is imported multiple times.

We would especially like to thank a large number of testers for pymbar
for helping us identify issues and ways to improve with the previous
pymbar release, including Tommy Knotts, David Mobley, Himanshu Paliwal,
Zhiqiang Tan, Patrick Varilly, Todd Gingrich, Aaron Keys, Anna Schneider,
Adrian Roitberg, Nick Schafer, Thomas Speck, Troy van Voorhis, Gupreet Singh,
Jason Wagoner, Gabriel Rocklin, Yannick Spill, Ilya Chorny, Greg Bowman, 
Vincent Voelz, Peter Kasson, Sam Moors, Carl Rogers, Josua Adelman, 
Javier Palacios, David Chandler, and Andrew Jewett.