- Timestamp:
- Dec 2, 2020 11:39:06 AM (2 years ago)
- File:
-
- 1 edited
Legend:
- Unmodified
- Added
- Removed
-
trunk/help/gsasII.html
r4664 r4668 3677 3677 3678 3678 <p class=MsoNormal style='margin-left:.5in'><span style='mso-fareast-font-family: 3679 "Times New Roman"'>This window provides the main controls and a few 3680 global parameters for GSAS-II. 3681 3682 <h5 style='margin-left:.5in'><span 3683 style='mso-fareast-font-family:"Times New Roman"'> 3684 What can I do here?</span></h5> 3685 <p style='margin-left:0.75in'>On this page, there are three or four sets of 3686 controls. The first is for how refinements operate. 3687 </p> 3679 "Times New Roman"'>This window provides access to the controls that 3680 determine how GSAS-II performs minimizations as well as few 3681 global parameters for GSAS-II. Note that many other customization 3682 settings are set as <A Href=#Preferences>configuration variables</a> 3683 in the Preferences menu. (See the 3684 <a 3685 href="https://gsas-ii.readthedocs.io/en/latest/GSASIIutil.html#module-config_example"> 3686 Programmer's documentation</a> for a description of those.) 3687 </P> 3688 3689 <p style='margin-left:0.75in'><U>Refinement Controls:</u><i> These 3690 controls determine how refinements are performed. The first determines 3691 the computational engine used to minimize the structure. 3692 </i></p> 3693 3688 3694 <A name="RefineType"></A> 3689 3695 <DL style='margin-left:1.0in'> … … 3693 3699 <UL style='margin-left:0.25in'> 3694 3700 <LI><U>analytic Hessian</U>: This is the default option and is usually 3695 the most useful. It uses a scustom-developed least-squares3701 the most useful. It uses a custom-developed least-squares 3696 3702 minimizer that uses singular-value decomposition (SVD) to reduce the 3697 3703 errors caused by correlated variables and the 3698 Levenberg-Marquardt algorithm to down-weight the off-diagonal3704 Levenberg-Marquardt algorithm to up-weight diagonal 3699 3705 Hessian terms when refinements fail to lower χ<sup>2</sup>. 3700 3706 <LI><u>analytic Jacobian</u>: This uses a numpy-provided <I>leastsq</I> 3701 minimizer, which not applicable for larger3702 problem s as it requires3707 minimizer, which not applicable for 3708 problem with a large number of histograms as it requires 3703 3709 much more memory than the Hessian routines. This because it 3704 creates a Jacobian matrix is shaped N x M (N parameters x M3705 observations) and uses that to create the N x N Hessian. The3706 "Hessian" minimizers create the Hessian matrix directly.3710 creates a Jacobian matrix that is shaped N x M (N parameters x M 3711 observations) while the Hessian methods create a Jacobian matrix 3712 only each histogram. 3707 3713 <LI><u>numeric</u>: This also uses the numpy <I>leastsq</I> 3708 3714 minimizer, and is also not applicable for larger 3709 problems. Unlike, the "analytic Jacobian", numerical derivat es3710 are computed for derivatives rather thananalytical derivatives3715 problems. Unlike, the "analytic Jacobian", numerical derivatives 3716 are computed rather than use the analytical derivatives 3711 3717 that are coded directly into GSAS-II. This will be slower than 3712 the analytical derivatives and will converge more slowly. It is 3718 the analytical derivatives and will is often less accurate which 3719 results in slower convergence. It is 3713 3720 typically used for code development to check the accuracy of the 3714 3721 analytical derivative formulations. … … 3716 3723 Hessian</U> but does not include the Levenberg-Marquardt 3717 3724 algorithm. It can be faster, but is more prone to 3718 diverge when severe correlation is present. 3725 diverge when severe correlation is present. It is possible that 3726 it might be better for single-crystal refinements. 3719 3727 </UL> 3720 3728 <p style='margin-left:0.25in'> … … 3735 3743 </p> 3736 3744 <DT><B>Min delta-M/M</B></DT> 3737 <DD>This determines when convergence is recognized; 3738 the refinement will stop when the change in the minimization 3739 function is less than this value. Set Min delta-M/M = 1.0 to force just a 3740 single cycle to be performed. The default is 0.001. A value less than 3741 10<sup>-4</sup> causes the refinement to cycle with no meaningful 3742 improvement since shifts become a small fraction of the 3743 parameter's uncertainties. Set Min delta-M/M = 1.0 to force 3744 refinement to stop after a single refinement. 3745 The allowed range is 10<sup>-9</sup> to 1.0. 3746 </DD> 3745 <DD>A refinement will stop when the change in the minimization function 3746 (M=Σ[w(Io-Ic)<sup>2</sup>]) 3747 is less than this value. The allowed range is 10<sup>-9</sup> to 3748 1.0, with a default of 0.001. A value of 3749 1.0 stops the refinement after a single cycle. Values less than 3750 10<sup>-4</sup> cause refinements to continue even if there is 3751 no meaningful improvement. 3752 </DD><P></P> 3747 3753 <DT><B>Max cycles</B></DT> 3748 3754 <DD>This determines the maximum number of refinement cycles that 3749 will be performed. This is only available with the "Hessian" minimizers.</DD> 3755 will be performed. This is only available with the "Hessian" minimizers.</DD><P></P> 3750 3756 <DT><B>Initial lambda</B></DT> 3751 <DD>Note that here λ is the Marquardt coefficient, which when large 3752 down-weights the significance of the off-diagonal terms in the 3757 <DD>Note that here λ is the Marquardt coefficient, where a weight of 3758 1+λ is applied to the diagonal elements of the Hessian. When λ is 3759 large, this down-weights the significance of the off-diagonal terms in the 3753 3760 Hessian. Thus, when λ is large, the refinement is effectively one of 3754 3761 steepest-descents, where correlation between variables is … … 3756 3763 slow and may not always find the local minimum. 3757 3764 This is only available with the "analytical Hessian" minimizer. 3758 </DD> 3765 </DD><P></P> 3759 3766 <DT><B>SVD zero tolerance</B></DT> 3760 3767 <DD>This determines the level where SVD considers values to be the 3761 3768 same. Default is 10<sup>-6</sup>. Make larger to where problems occur due to correlation. This is only available with the "Hessian" minimizers. 3762 </DD> 3769 </DD><P></P> 3763 3770 <DT><B>Initial shift factor</B></DT> 3764 <DD>? 3765 </DD> 3771 <DD>A “damping multiplier” applied during the first refinement 3772 cycle, for Jacobean/numeric refinements only. 3773 Should be in interval (0.1, 100). See the 3774 <A href="https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.leastsq.html">SciPy leastsq docs</A> 3775 for more information. 3776 </DD><P></P> 3766 3777 </DL> 3767 3778 <A name="SingleXtlSettings"></A> 3768 <p style='margin-left:0.75in'> A set of controls is provided for3779 <p style='margin-left:0.75in'><U>Single Crystal:</u><i> A set of controls is provided for 3769 3780 control of single-crystal refinements. 3770 3781 These only appear when single crystal (HKLF) histograms are present 3771 in the project. 3782 in the project. </i> 3772 3783 </p> 3773 3784 <DL style='margin-left:1.0in'> … … 3775 3786 <DD>When checked, refinements are against F<sup>2</sup> rather than 3776 3787 |F|. 3777 </DD> 3788 </DD><P></P> 3778 3789 <DT><B>Min obs/sig</B></DT> 3779 <DD>? 3780 </DD> 3790 <DD>Conventional cutoff for single crystal refinements as to what reflections 3791 should be considered observed, typical values are 2.0 (2σ) or 3.0 (3σ). 3792 </DD><P></P> 3781 3793 <DT><B>Min extinct.</B></DT> 3782 <DD> ?3783 </DD> 3794 <DD>(needs further work) 3795 </DD><P></P> 3784 3796 <DT><B>Max delt-F/sig</B></DT> 3785 <DD>? 3786 </DD> 3797 <DD>Removes reflections that are very poorly fit. Should be used 3798 only with extreme care, since poorly-fit reflections could be an 3799 indication that the structure is wrong. 3800 </DD><P></P> 3787 3801 <DT><B>Max d-spacing</B></DT> 3788 3802 <DD>Reflections with d-space values larger than this value are ignored. 3789 </DD> 3803 </DD><P></P> 3790 3804 <DT><B>Min d-spacing</B></DT> 3791 3805 <DD>Reflections with d-space values smaller than this value are ignored. 3792 </DD> 3806 </DD><P></P> 3793 3807 3794 3808 </DL> 3795 3809 3796 3810 <A name="SequentialSettings"></A> 3797 <p style='margin-left:0.75in'>A set of controls is for 3798 sequential refinement. Settings here determine if "normal" or "sequential" 3799 refinement is performed. If no datasets are selected, then all "used" 3800 histograms are included in one combined refinement. However, if any 3801 number histogram are selected used here, then a 3802 sequential refinement is performed, where a fit is made to each 3803 histogram in turn. Only the first item below is shown in "normal" mode. 3811 <p style='margin-left:0.75in'><U>Sequential Settings:</u><i> 3812 A set of controls is for 3813 sequential refinement. Settings here determine if a "normal" or "sequential" 3814 refinement is performed. If no datasets are selected here, then all 3815 histograms linked to phases in the project and that are flagged as 3816 "used" are included in one potentially large (combined) 3817 refinement. However, if any 3818 number of histograms are selected here, then a 3819 sequential refinement is performed, where a fit is made to the 3820 structural model(s) fitting each selected 3821 histogram in turn. Only the first item below is shown in "normal" mode. </i> 3804 3822 </p> 3805 3823 <DL style='margin-left:1.0in'> … … 3811 3829 set as "normal". The button is labeled "Select" when in normal refinement 3812 3830 mode and "Reselect" in sequential refinement mode. 3813 </DD> 3831 </DD><P></P> 3814 3832 <DT><B>Reverse order?</B></DT> 3815 3833 <DD>Normally, in a sequential histograms are fit in the order they … … 3818 3836 but when this option is selected, the sequential fit is performed 3819 3837 with the last tree entry first. 3820 </DD> 3838 </DD><P></P> 3821 3839 <DT><B>Copy results to next histogram?</B></DT> 3822 3840 <DD>When this option is selected, the fitted parameters from each … … 3830 3848 refinement is completed. For subsequent refinements, it is usually 3831 3849 better to start with the results from the previous fit. 3832 </DD> 3850 </DD><P></P> 3833 3851 <DT><B>Clear previous seq. results</B></DT> 3834 3852 <DD>When this button is pressed, the "Sequential Results" entry 3835 3853 with the results from the last sequential fit is deleted from the 3836 3854 tree. 3837 </DD> 3855 </DD><P></P> 3838 3856 </DL> 3857 <p style='margin-left:0.75in'><U>Global Settings:</u><i> 3858 This is a location for parameters that apply to an entire project. 3859 At present there is only one. </i> 3860 </p> 3861 <DL style='margin-left:1.0in'> 3862 <DT><B>CIF Author</B></DT> 3863 <DD>The value provided here is used when creating a CIF of an 3864 entire project. 3865 </DD><P></P> 3866 </DL> 3867 3868 <h5 style='margin-left:.5in'><span 3869 style='mso-fareast-font-family:"Times New Roman"'> 3870 What can I do here?</span></h5> 3871 <p style='margin-left:0.75in'>This offers a place to change how 3872 GSAS-II performs refinements, but has no specific menu commands or 3873 graphics. 3874 </p> 3839 3875 3840 3876 <h4 style='margin-left:0.25in'><a … … 7532 7568 <hr size=2 width="100%" align=center> 7533 7569 7534 <!-- hhmts start -->Last modified: Sun Nov 29 20:05:28CST 2020 <!-- hhmts end -->7570 <!-- hhmts start -->Last modified: Wed Dec 2 11:04:42 CST 2020 <!-- hhmts end --> 7535 7571 7536 7572 </div>
Note: See TracChangeset
for help on using the changeset viewer.