
Each derivative should be carefully monitored for evidence of nonisomorphism,
as this is a major source of difficulties during refinement; any changes in cell lengths
relative to the native crystal should be at most
» 0.5% (this will give
» 15% mean
intensity change at 3Å resolution). A plot (e.g. using SCALEIT) of mean isomorphous
difference vs. resolution should decrease smoothly at the high resolution end. Note
that if nonisomorphism is detected, it is not sufficient simply to apply a high
resolution cutoff at the point of upturn in the plot; ideally the dataset should be
discarded completely.
From the isomorphous difference Patterson, locate 1 or, even better, 2 major sites in
your "best" derivative. Frequently what happens is that you collect data for many
derivatives, but none of the Pattersons appear readily soluble. However, provided
a derivative is not nonisomorphous it may still be usable for phasing. Then one day,
the next Patterson is soluble: this is your best derivative! At this point you stop
collecting data and start refining.

Do 5 cycles of phasing and refinement of the coordinates and site occupancies
(initially set to 1) for this derivative. At this stage it is advisable to keep the overall
scale and isotropic thermal parameter fixed (at the default values of 1 and 0 resp.),
to fix the individual thermal parameters (e.g. 25), and to use a high resolution data
cutoff (e.g. if the derivative data extends to 3Å, cutoff at 4.5Å). Also the anomalous
data should not be included yet, and centric data only should be used in the
refinement, provided the space group has more than one centric zone. Note that
MLPHARE doesn't need initial estimates of RMS lack of closure errors.
#
mlphare HKLIN pbgd_fhscal HKLOUT ufmr <<EOD
CYCLES 5
LABIN FP=FNAT SIGFP=SIGFNAT 
FPH1=FUF SIGFPH1=SIGFUF
LABOUT ALLIN
PRINT AVF AVE
THRESH 2.5 .5
RESOL 20 4.5
CENTRI
DERIV UF
DCYCLE PHASE ALL REFCYC ALL
ATOM U 0.28 0.18 0.16 1 BFAC 25
ATREF X ALL Y ALL Z ALL OCC ALL
EOD

Using the reflection file output by MLPHARE, calculate a difference Fourier with
the FFT program, and run a peak search program (e.g. PEAKMAX) on it, listing the
highest 10 peaks. Examine the peak listing, and eliminate any symmetryrelated
peaks at the edges of the map. You will certainly see the sites that were input (even
if they are wrong!  but at this stage you have to assume they are right). Take the
remaining peaks in order one at a time, and check against the Patterson (e.g. with
program RSPS or VECSUM), until a site is found or the list is exhausted.
#
fft HKLIN ufmr MAPOUT ufmrdf <<EOD
TITLE UF diff Fourier phased on UF derivative refined by MLPHARE.
LABIN F1=FUF SIG1=SIGFUF F2=FNAT SIG2=SIGFNAT PHI=PHIB W=FOM
EOD
if ($status) exit
peakmax MAPIN ufmrdf <<EOD
NUMPEA 10
OUTPUT NONE
EOD
rm ufmrdf.map

Add any new site found to the input for MLPHARE, and repeat steps b and c,
until no new sites are found. The statistics for each derivative output by MLPHARE
on the final phasing cycle should always be checked. A reduction in R_{Cullis} (average
Pweighted lack of closure divided by average isomorphous difference) is supposed
to be a good validator of a new site.
#
mlphare HKLIN pbgd_fhscal HKLOUT ufmr <<EOD
CYCLES 5
LABIN FP=FNAT SIGFP=SIGFNAT 
FPH1=FUF SIGFPH1=SIGFUF
LABOUT ALLIN
PRINT AVF AVE
THRESH 2.5 .5
RESOL 20 4.5
CENTRI
DERIV UF
DCYCLE PHASE ALL REFCYC ALL
ATOM U 0.282 0.173 0.158 0.889 BFAC 25
ATREF X ALL Y ALL Z ALL OCC ALL
ATOM U 0.18731 0.17452 0.11287 .5 BFAC 25
ATREF X ALL Y ALL Z ALL OCC ALL
EOD
if ($status) exit
fft HKLIN ufmr MAPOUT ufmrdf <<EOD
TITLE UF diff Fourier phased on UF derivative refined by MLPHARE.
LABIN F1=FUF SIG1=SIGFUF F2=FNAT SIG2=SIGFNAT PHI=PHIB W=FOM
EOD
if ($status) exit
peakmax MAPIN ufmrdf <<EOD
NUMPEA 10
OUTPUT NONE
EOD
rm ufmr.mtz ufmrdf.map

Repeat step c, but this time calculate a crossdifference Fourier on another
derivative.
#
fft HKLIN ufmr MAPOUT uamrdf <<EOD
TITLE UAc diff Fourier phased on UF derivative refined by MLPHARE.
LABIN F1=FUAC SIG1=SIGFUAC F2=FNAT SIG2=SIGFNAT PHI=PHIB W=FOM
EOD
if ($status) exit
peakmax MAPIN uamrdf <<EOD
NUMPEA 10
OUTPUT NONE
EOD
rm ufmr.mtz uamrdf.map

Add the new derivative and its site to the input for MLPHARE and repeat steps
b, c and d on the new derivative.
#
mlphare HKLIN pbgd_fhscal HKLOUT uamr <<EOD
CYCLES 5
LABIN FP=FNAT SIGFP=SIGFNAT 
FPH1=FUF SIGFPH1=SIGFUF 
FPH2=FUAC SIGFPH2=SIGFUAC
LABOUT ALLIN
PRINT AVF AVE
THRESH 2.5 .5
RESOL 20 4.5
CENTRI
DERIV UF
DCYCLE PHASE ALL REFCYC ALL
ATOM U 0.284 0.166 0.162 0.926 BFAC 25.000
ATREF X ALL Y ALL Z ALL OCC ALL
ATOM U 0.185 0.176 0.106 0.603 BFAC 25.000
ATREF X ALL Y ALL Z ALL OCC ALL
ATOM U 0.499 0.243 0.381 0.281 BFAC 25.000
ATREF X ALL Y ALL Z ALL OCC ALL
DERIV UAc
DCYCLE PHASE ALL REFCYC ALL
ATOM U 0.284 0.170 0.160 1 BFAC 25.000
ATREF X ALL Y ALL Z ALL OCC ALL
EOD
if ($status) exit
fft HKLIN uamr MAPOUT uamrdf <<EOD
TITLE UAc diff Fourier phased on UF & UAc derivatives refined by MLPHARE.
LABIN F1=FUAC SIG1=SIGFUAC F2=FNAT SIG2=SIGFNAT PHI=PHIB W=FOM
EOD
if ($status) exit
peakmax MAPIN uamrdf <<EOD
NUMPEA 10
OUTPUT NONE
EOD
rm uamr.mtz uamrdf.map

Repeat steps e and f for all remaining derivatives (at this stage no sites consistent
with the Patterson could be found for the PCMBS derivative, so it was not included).

At this point any anomalous data should be brought into play, and it is first
necessary to establish the "hand" of the heavy atoms. The easiest way to do this is
to set all "anomalous occupancies" to zero, and using both centric and acentric data,
do 10 cycles of phasing and refinement. The high resolution cutoff should be
removed.
The anomalous occupancies should all refine positive if the hand is already correct,
or negative if it is wrong, in which case the coordinates of all sites must be inverted
(i.e. x,y,z becomes x,y,z , which is equivalent to x,y,z in point group 222).
If there is no consistent change in the anomalous occupancies, the anomalous data
should be left out and brought in later; alternatively it may be possible to include
anomalous data for only the higher occupancy derivatives.
#
mlphare HKLIN pbgd_fhscal HKLOUT allmr1 <<EOD
TITLE Refine 5 derivatives with anomalous occupancy = 0.
CYCLES 10
LABIN FP=FNAT SIGFP=SIGFNAT 
FPH1=FUF SIGFPH1=SIGFUF DPH1=DANUF SIGDPH1=SIGDANUF 
FPH2=FUAC SIGFPH2=SIGFUAC DPH2=DANUAC SIGDPH2=SIGDANUAC 
FPH3=FUS SIGFPH3=SIGFUS 
FPH4=FPTCL SIGFPH4=SIGFPTCL DPH4=DANPTCL SIGDPH4=SIGDANPTCL 
FPH5=FYBCL SIGFPH5=SIGFYBCL
LABOUT ALLIN
PRINT AVF AVE
THRESH 2.5 .5
DERIV UF
DCYCLE PHASE ALL REFCYC ALL
ATOM U 0.287 0.171 0.163 0.968 0 BFAC 25
ATREF X ALL Y ALL Z ALL OCC ALL AOCC ALL
ATOM U 0.495 0.238 0.380 0.547 0 BFAC 25
ATREF X ALL Y ALL Z ALL OCC ALL AOCC ALL
ATOM U 0.186 0.187 0.109 0.615 0 BFAC 25
ATREF X ALL Y ALL Z ALL OCC ALL AOCC ALL
ATOM U 0.986 0.051 0.510 0.314 0 BFAC 25
ATREF X ALL Y ALL Z ALL OCC ALL AOCC ALL
DERIV UAc
DCYCLE PHASE ALL REFCYC ALL
ATOM U 0.188 0.186 0.110 0.974 0 BFAC 25
ATREF X ALL Y ALL Z ALL OCC ALL AOCC ALL
ATOM U 0.292 0.170 0.155 0.837 0 BFAC 25
ATREF X ALL Y ALL Z ALL OCC ALL AOCC ALL
ATOM U 0.486 0.235 0.370 0.829 0 BFAC 25
ATREF X ALL Y ALL Z ALL OCC ALL AOCC ALL
DERIV US
DCYCLE PHASE ALL REFCYC ALL
ATOM U 0.180 0.195 0.111 0.625 BFAC 25
ATREF X ALL Y ALL Z ALL OCC ALL
DERIV Pt
DCYCLE PHASE ALL REFCYC ALL
ATOM PT 0.254 0.042 0.589 0.910 0 BFAC 25
ATREF X ALL Y ALL Z ALL OCC ALL AOCC ALL
DERIV Yb
DCYCLE PHASE ALL REFCYC ALL
ATOM YB 0.493 0.234 0.381 0.413 BFAC 25
ATREF X ALL Y ALL Z ALL OCC ALL

Using the newly determined phases, repeat the difference Fourier and peak search
for each derivative, adding any new sites found. Repeat the phasing and refinement,
including the overall scale and thermal parameter, and possibly the individual
thermal parameters, though it is often found that these tend to be unstable.
#
mlphare HKLIN pbgd_fhscal HKLOUT allmr2 <<EOD
TITLE Refine all derivatives (hand inverted with z changed to 1z).
CYCLES 10
LABIN FP=FNAT SIGFP=SIGFNAT 
FPH1=FUF SIGFPH1=SIGFUF DPH1=DANUF SIGDPH1=SIGDANUF 
FPH2=FUAC SIGFPH2=SIGFUAC DPH2=DANUAC SIGDPH2=SIGDANUAC 
FPH3=FUS SIGFPH3=SIGFUS 
FPH4=FPTCL SIGFPH4=SIGFPTCL DPH4=DANPTCL SIGDPH4=SIGDANPTCL 
FPH5=FYBCL SIGFPH5=SIGFYBCL 
FPH6=FPCMBS SIGFPH6=SIGFPCMBS DPH6=DANPCMBS SIGDPH6=SIGDANPCMBS
LABOUT ALLIN
PRINT AVF AVE
THRESH 2.5 .5
DERIV UF
DCYCLE PHASE ALL REFCYC ALL KBOV ALL
ATOM U 0.287 0.172 0.837 0.811 0.638 BFAC 13.881
ATREF X ALL Y ALL Z ALL OCC ALL AOCC ALL B ALL
ATOM U 0.494 0.240 0.619 0.564 0.425 BFAC 12.593
ATREF X ALL Y ALL Z ALL OCC ALL AOCC ALL B ALL
ATOM U 0.185 0.187 0.892 0.607 0.476 BFAC 15.953
ATREF X ALL Y ALL Z ALL OCC ALL AOCC ALL B ALL
ATOM U 0.988 0.052 0.489 0.437 0.460 BFAC 65.797
ATREF X ALL Y ALL Z ALL OCC ALL AOCC ALL B ALL
DERIV UAc
DCYCLE PHASE ALL REFCYC ALL KBOV ALL
ATOM U 0.186 0.186 0.890 0.912 0.759 BFAC 19.199
ATREF X ALL Y ALL Z ALL OCC ALL AOCC ALL B ALL
ATOM U 0.289 0.172 0.845 0.732 0.537 BFAC 16.749
ATREF X ALL Y ALL Z ALL OCC ALL AOCC ALL B ALL
ATOM U 0.488 0.239 0.627 0.755 0.593 BFAC 17.873
ATREF X ALL Y ALL Z ALL OCC ALL AOCC ALL B ALL
DERIV US
DCYCLE PHASE ALL REFCYC ALL KBOV ALL
ATOM U 0.181 0.195 0.889 0.571 BFAC 34.773
ATREF X ALL Y ALL Z ALL OCC ALL B ALL
DERIV Pt
DCYCLE PHASE ALL REFCYC ALL KBOV ALL
ATOM PT 0.252 0.044 0.409 0.878 0.871 BFAC 33.104
ATREF X ALL Y ALL Z ALL OCC ALL AOCC ALL B ALL
DERIV Yb
DCYCLE PHASE ALL REFCYC ALL KBOV ALL
ATOM YB 0.493 0.237 0.619 0.458 BFAC 19.003
ATREF X ALL Y ALL Z ALL OCC ALL B ALL
DERIV PCMBS
DCYCLE PHASE ALL REFCYC ALL KBOV ALL
ATOM HG 0.245 0.060 0.145 0.271 0.252 BFAC 52.205
ATREF X ALL Y ALL Z ALL OCC ALL AOCC ALL B ALL
ATOM HG 0.074 0.064 0.154 0.326 0.311 BFAC 47.182
ATREF X ALL Y ALL Z ALL OCC ALL AOCC ALL B ALL
EOD

Repeat step i until there is no further change in the list of sites. Note that once a
good derivative is wellrefined and there are obviously no new sites to be found, its
refinement flags can be switched off, and refinement performed on only the weaker
derivatives. The printed "refinement parameters" indicate the progress of
convergence of refinement for each derivative.
Note that in the above procedure, only the Patterson for the first (and best) derivative
needs to be solved; the other derivatives are solved from the difference Fouriers, and
the Pattersons, which are often difficult to solve ab initio, are then only used to crosscheck the new sites. This also obviates the problem of ensuring that all derivatives
are solved relative to the same origin and on the same hand. Of course, if more than
one Patterson can be solved independently, so much the better, but then difference
Fouriers must still be used to correlate the origins and the hand.