Heisenbug in Topas v5? exp_conv_const and dummy_str

rowlesmr

Can anyone please try this input file (in any version)?

It runs less than half the time on my installation of Topas 5. It will refine to completion or it will fail on the first iteration with a "Unhandled exception in derivatives - Maybe parameter limits would help" error message.

I currently don't have access to my other computer.

The cause (I think) is the interaction between the exp_conv_const (absorption correction) and dummy_str; if I remove any one of them, the problem goes away.


yobs_eqn row119.xy = 1; min 6 max 94 del 0.02
	r_wp  24.8113986 r_exp  1.49076319 gof  16.6434204
	bkg  @  1363.21829`  933.126465` -796.412984`  392.453642`
	
	local abs  64.43478`_2.59830 min 5 max 500
	exp_conv_const = Ln(0.001) Sin(2 Th) / ( (abs) Rs 0.003490658 );
 	
	Rp 200
	Rs 200	
	
   lam 
   	ymin_on_ymax 0.0001         
   	la 1  lo 1.7889847	lh 0.4633522  

	dummy_str 
		phase_name "Dummy str for MAC calc"
		prm mac_prm = Get(phase_MAC); :  54.66423
		scale 0
		space_group P1 a 5 b 5 c 5
		al 90 be 90 ga 90
		site q x 0 y 0 z 0 
			occ Si  =50/28.086; :  1.78025
			occ O  =50/15.999; :  3.12520             

	str
		phase_name "Quartz"
		scale @ 0.000038
		a 4.9134
		b = Get(a);
		c 5.4052
		ga 120
		space_group "P 32 2 1"
		site Si1 num_posns  3	x 0.46987 y 0 z =2/3; occ Si+4 1. beq 0.466
		site O1 num_posns  6	x 0.4141 y 0.2681 z 0.7855 occ O-2 1. beq 1.026

johnsoevans

Dear Matthew,

I didn't look at the file, but it ran ok every time on ta.exe v7.19 (one behind today's latest v7.20). It also ran ok on ta.exe v6 (not sure what version, date stamp 29/7/2017).

Best,

John

rowlesmr

Ta a lot. I'll have another look when I get home. Hopefully another arrow in my quiver to get work to upgrade.

And in other news, a new personal best:

29 phases.

alancoelho

V6 and V7 seems to work.

V5 seems to have a problem with dummy_str together with exp_conv_const.

A work around is to place the exp_conv_const at str level which then works.

rowlesmr

Yep.

Works in my version 6 and v7.20 at home. Same behaviour in my version 5.

.

That my also explain some other flaky behaviour I had with some other dummy_str things I've used previously. I may have to experiment with putting the absorption correction in a for strs{} and for hkl_iss{}

.

In other, other news, it took ~0.2 s to finish in v6 and ~0.08 s in v7. When it did work in v5, it took ~0.2-0.4 s. I tested first in v7, and it was noticeably slower in 6. Thanks, Alan!

rowlesmr

Thanks Alan.

Moved my Absorption correction to a for strs loop and it looks like it works.

for xdds {
	if Prm_There(startX) {} else { start_X 5 }
	Specimen_Displacement(sd)
	Zero_Error(ze)
	
	 'if the absorption loops go here, it doesn't work.
	
#ifndef COBALT
	LP_Factor_Perpendicular(26.55)

	Rp 200
	Rs 200
	
	axial_conv 
		filament_length  12
		sample_length  20
		receiving_slit_length  14
		primary_soller_angle    !sol  3.14229_0.02043 '=0.04 Rad; :   0
		secondary_soller_angle  !sol  3.14229_0.02043 '=0.04 Rad; :   0 
		axial_n_beta  30
	lpsd_th2_angular_range_degrees  4
		lpsd_equitorial_divergence_degrees  0.125
	
	lam
		ymin_on_ymax  0.0001
		la  0.0159 lo  1.534753 lh  3.6854
		la  0.5791 lo  1.540596 lh  0.437 
		la  0.0762 lo  1.541058 lh  0.6 
		la  0.2417 lo  1.54441  lh  0.52
		la  0.0871 lo  1.544721 lh  0.62
		la  .02 lo   0.7693222_0.000876436368 lg   35 max 70 ' Cu ka / 2
		la  0.00494_0.00028 lo  =12.398/8.3976; :  1.4763742 lh  0.437 ' W La1.4763742 
		la  0.00340_0.00028	 lo 1.3922160	lh 0.5502872 ' Cu kB
#else

	LP_Factor( 2.68)
	
	Rp 200
	Rs 200	
	
	axial_conv 
		filament_length  12
		sample_length  10
		receiving_slit_length  14
		primary_soller_angle   !sol  3.75708 ' =0.04 Rad; :   2.29183
		secondary_soller_angle !sol  3.75708 ' =0.04 Rad; :   2.29183 
		axial_n_beta  30
	lpsd_th2_angular_range_degrees  4
		lpsd_equitorial_divergence_degrees 0.125

	
		
	prm !co_kB_sc  0.00328 min 0.0001
   lam 
   	ymin_on_ymax 0.0001 
      'ka
   	la 0.378  lo 1.7889847	lh 0.4633522  
      la 0.144  lo 1.7892524	lh 0.6958819 
      la 0.127  lo 1.7896946	lh 1.1767380 
      la 0.088  lo 1.7888515	lh 0.2085420 
      la 0.197  lo 1.7927905	lh 0.6237179 
      la 0.095  lo 1.7930637	lh 0.7190761 
      la 0.050  lo 1.7934738	lh 1.1578452 
      'kb
      la =0.449 co_kB_sc; :  0.00147	 lo 1.6207938	lh 0.6462327


#endif
		
	if Obj_There(str) { for strs { Absorption(abs) } }
	if Obj_There(hkl_Is) { for hklis { Absorption(abs) } }
	
		
}