relax 3.3.0

From relax wiki
Jump to: navigation, search

Official relax releases
relax logo
relax version 3.3.0
Previous version Next version
← relax 3.2.3 relax 3.3.1 →

Keywords Relaxation dispersion, speed
Release type Major feature
Release date 3 September 2014

The PDF version of the relax 3.3.0 user manual The relax 3.3.0 user manual


This is a major feature release which includes a huge number of changes, as can be seen below. The most important change is an incredible speed up of all relaxation dispersion models. See the table below for a comparison to the previous relax 3.2.3 release. The maximum possible advantage of linear algebra operations are used to eliminate all of the slow Python looping and to obtain the ultimate algorithms for speed. As this is using NumPy, conversion to C or FORTRAN will not result in any significant speed advantage. With these huge speed ups, relax should now be one of the fastest software packages for analysing relaxation dispersion phenomena.

Other important features include the implementation of a zooming grid search algorithm for use in all analysis types, expanded plotting capabilities for R values in the relaxation dispersion analysis, the ability to optimise the R1 parameter in all off-resonance dispersion models, proper minimisation statistics resetting by the minimisation user functions, and a large expansion of the periodic table information for all elements in the relax library for correctly estimating molecular masses. Additional features are that there is better tab completion support in the prompt UI for Mac OS X, the addition of the time user function for printing the current date and time, the value.copy user function accepting a force argument for overwriting values, model nesting in the dispersion auto-analysis has been extended, the spin-lock offset is now shown in the dispersion analysis in the GUI, the relax_disp.r2eff_estimate user function has been added for fast R2eff and I0 parameter value and error estimation, and gradient and Hessian functions have been added to the exponential curve-fitting C module allowing for more advanced optimisation in the relaxation curve-fitting and dispersion analyses.

Note that this new 3.3 relax series breaks compatibility with old relax scripts. The important change, which is the main reason for starting the relax 3.3.x line, is the renaming of the calc, grid_search and minimise user functions to minimise.calculate, minimise.grid_search and minimise.execute respectively. Please update your scripts appropriately. A new relax feature is that old user function calls are detected in the prompt and script UIs and a RelaxError raised explaining what to rename the user function to.

Important bugfixes in this release include that relax can run on MS Windows systems again, numerous Python 3 fixes, the ability to load Bruker DC files when the file format has corrupted whitespace, the GUI "close all analyses" feature works and no longer raises an error, structure.create_diff_tensor_pdb user function now works when no structural data is present, the geometric prolate diffusion 3D PDB representation in a model-free analysis now aligns with the axis in the PDB as it was previously rotated by 90 degrees, and the Monte Carlo simulations in the relaxation dispersion analysis for exponential curve-fitting for R2eff/R parameter errors is now correct and no longer underestimating the errors by half. For more details about the new features and the bug fixes, please see below. For fully formatted and easy to navigate release notes, please see

To demonstrate the huge speeds ups in the relaxation dispersion analysis, the following table compares the speed of dispersion models in relax 3.2.3 compared to the new 3.3.0 version:

100 single spins analysis (times in seconds):
Dispersion model relax 3.2.3 timings relax 3.3.0 timings Speed change
No Rex 0.824±0.017 0.269±0.016 3.068x faster.
LM63 1.616±0.017 0.749±0.008 2.157x faster.
LM63 3-site 3.218±0.039 0.996±0.013 3.230x faster.
CR72 2.639±0.042 1.536±0.019 1.718x faster.
CR72 full 2.808±0.027 1.689±0.075 1.663x faster.
IT99 1.838±0.032 0.868±0.011 2.118x faster.
TSMFK01 1.643±0.033 0.718±0.011 2.289x faster.
B14 5.841±0.050 3.747±0.044 1.559x faster.
B14 full 5.942±0.053 3.841±0.044 1.547x faster.
NS CPMG 2-site expanded 8.309±0.066 4.070±0.073 2.041x faster.
NS CPMG 2-site 3D 245.180±2.162 45.410±0.399 5.399x faster.
NS CPMG 2-site 3D full 237.217±2.582 45.177±0.415 5.251x faster.
NS CPMG 2-site star 183.423±1.966 36.542±0.451 5.020x faster.
NS CPMG 2-site star full 183.622±1.326 36.788±0.343 4.991x faster.
MMQ CR72 5.920±0.105 4.078±0.105 1.452x faster.
NS MMQ 2-site 363.659±2.610 82.588±1.197 4.403x faster.
NS MMQ 3-site linear 386.798±4.480 92.060±0.754 4.202x faster.
NS MMQ 3-site 391.195±3.442 93.025±0.829 4.205x faster.
M61 1.576±0.022 0.862±0.009 1.828x faster.
DPL94 22.794±0.517 1.101±0.008 20.705x faster.
TP02 19.892±0.363 1.232±0.007 16.152x faster.
TAP03 31.701±0.378 1.936±0.017 16.377x faster.
MP05 24.918±0.572 1.428±0.015 17.454x faster.
NS R1rho 2-site 244.604±2.493 35.125±0.202 6.964x faster.
NS R1rho 3-site linear 287.181±2.939 68.245±0.536 4.208x faster.
NS R1rho 3-site 290.486±3.614 70.449±0.686 4.123x faster.
Cluster of 100 spins analysis (times in seconds):
Dispersion model relax 3.2.3 timings relax 3.3.0 timings Speed change
No Rex 0.818±0.016 0.008±0.001 97.333x faster.
LM63 1.593±0.018 0.037±0.000 43.401x faster.
LM63 3-site 3.134±0.039 0.067±0.001 47.128x faster.
CR72 2.610±0.047 0.115±0.001 22.732x faster.
CR72 full 2.679±0.034 0.122±0.005 22.017x faster.
IT99 1.807±0.025 0.063±0.001 28.687x faster.
TSMFK01 1.636±0.036 0.039±0.001 42.170x faster.
B14 5.799±0.054 0.488±0.010 11.879x faster.
B14 full 5.803±0.043 0.484±0.006 11.990x faster.
NS CPMG 2-site expanded 8.326±0.081 0.685±0.012 12.160x faster.
NS CPMG 2-site 3D 244.869±2.382 41.217±0.467 5.941x faster.
NS CPMG 2-site 3D full 236.760±2.575 41.001±0.466 5.775x faster.
NS CPMG 2-site star 183.786±2.089 30.896±0.417 5.948x faster.
NS CPMG 2-site star full 183.243±1.615 30.898±0.343 5.931x faster.
MMQ CR72 5.978±0.094 0.847±0.007 7.061x faster.
NS MMQ 2-site 363.138±3.041 75.906±0.845 4.784x faster.
NS MMQ 3-site linear 384.978±5.402 83.703±0.773 4.599x faster.
NS MMQ 3-site 388.557±3.261 84.702±0.762 4.587x faster.
M61 1.555±0.021 0.034±0.001 45.335x faster.
DPL94 22.837±0.494 0.140±0.002 163.004x faster.
TP02 19.958±0.407 0.167±0.002 119.222x faster.
TAP03 31.698±0.424 0.287±0.003 110.484x faster.
MP05 25.009±0.683 0.187±0.007 133.953x faster.
NS R1rho 2-site 242.096±1.483 32.043±0.157 7.555x faster.
NS R1rho 3-site linear 280.778±2.589 62.866±0.616 4.466x faster.
NS R1rho 3-site 282.192±5.195 63.174±0.816 4.467x faster.

Full details of this comparison can be seen in the test_suite/shared_data/dispersion/profiling directory. For information about each of these models, please see the links:,,,,,,,,,,,,,,,,,,,,,,,,,

For CPMG statistics: 3 fields, each with 20 CPMG points. Total number of dispersion points per spin is 60.

For R experiments: 3 fields, each with 10 spin lock offsets, and each offset has been measured at 5 different spin lock fields. Per field there is 50 dispersion points. Total number of dispersion points per spin is 150.


The new relax versions can be downloaded from If binary distributions are not yet available for your platform and you manage to compile the binary modules, please consider contributing these to the relax project (described in section 3.6 of the relax manual,


Version 3.3.0
(3 September 2014, from /trunk)


  • Huge speed ups for all of the relaxation dispersion models ranging from 1.452x to 163.004x times faster. The speed ups for the clustered spin analysis are far greater than for the single spin analysis.
  • Implementation of a zooming grid search algorithm for optimisation in all analyses. This includes the addition of the minimise.grid_zoom user function to set the zoom level. The grid width will be divided by 2zoom_level and centred at the current parameter values. If the new grid is outside of the bounds of the original grid, the entire grid will be translated so that it lies entirely within the original.
  • Increased the amount of user feedback for the minimise.grid_search user function. Now a comment for each parameter is included in the printed grid search setup table. This includes if the lower or upper bounds, or both, have been supplied and if a preset value has been used instead.
  • Expanded support for R 2D graph plotting in the relax_disp.plot_disp_curves user function as the X-axis can now be the ν1 value, the effective field ωeff, or the rotating frame title angle θ. And the plots are interpolation over the spin-lock offset.
  • Ability to optimise the R1 relaxation rate parameter in the off-resonance relaxation dispersion models.
  • Creation of the relax_disp.r1_fit user function for activating and deactivating R1 fitting in the dispersion analysis.
  • Better tab completion support in the prompt UI for Mac OS X users. For some Python versions, the Mac supplied libedit library is used rather than GNU readline. But this library uses a completely different language and hence tab completion was non-functional on these systems. The library difference is now detected and the correct language sent into libedit to activate tab completion.
  • Created the time user function. This is just a shortcut for printing out the output of the time.asctime() function.
  • The value.copy user function now accepts the force flag to allow destination values to be overwritten.
  • Expanded model nesting capabilities in the relaxation dispersion auto-analysis to speed up the protocol.
  • The spin-lock offset is now included in the spectra list GUI element for the relaxation dispersion analysis.
  • Creation of the relax_disp.r2eff_estimate user function for the fast estimation of R2eff/R values and errors when full exponential curves have been collected. This experimental feature uses linearisation to estimate the R2eff and I0 parameters and the covariance matrix to estimate parameter errors.
  • Gradients and Hessians are implemented for the exponential curve-fitting, hence all optimisation algorithms and constraint algorithms are now available for this analysis type. Using Newton optimisation instead of Nelder-Mead Simplex can save over an order of magnitude in computation time. This is also available in the relaxation dispersion analysis.
  • The minimisation statistics are now being reset for all analysis types. The minimise.calculate, minimise.grid_search, and minimise.execute user functions now all reset the minimisation statistics for either the model or the Monte Carlo simulations prior to performing any optimisation. This is required for both parallelised grid searches and repetitive optimisation schemes to allow the result to overwrite an old result in all situations, as sometimes the original chi-squared value is lower and the new result hence is rejected.
  • Large expansion of the periodic table information in the relax library to include all elements, the IUPAC 2011 standard atomic weights for all elements, mass numbers and atomic masses for all stable isotopes, and gyromagnetic ratios.
  • Significant improvements to the structure centre of mass calculations by using the new periodic table information - all elements are now supported and exact masses are now used.
  • Added a button to the spectra list GUI element for the spectrum.error_analysis user function. This is placed after the 'Add' and 'Delete' buttons and is used in the NOE, R1 and R2 curve-fitting and relaxation dispersion analyses.
  • RelaxErrors are now raised in the prompt or script UI if an old user function is called, printing out the names of the old and new user functions. This is for help in upgrading old scripts and is currently for the calc(), grid_search(), and minimise() user function calls.




For reference, the announcement for this release can also be found at following links:

Softpedia also has information about the newest relax releases:


If you would like to receive announcements about new relax versions, please subscribe to the relax-announce mailing list using the relax-announce info page. This list only receives ~10 emails per year. It is archived at Gmane (thread, blog, NNTP, RSS), in the local archives, in The Mail Archive, and in the MARC (Mailing list ARChives).

See also