--------------------------------------------------------------- Changes in input files for year 2000 version of gato change names: disfac => dsplnrm scdflt => psiscal normal values: dsplnrm = 0.10 psiscal = 0.40 Restart and maximum decomposition control: nrestrt = 1 => restart from info in the rgta file. mxdcomp = Maximum number of decompositions to perform. Stop when this maximum is reached. nitmax = 0 => No inverse iterations performed. Up/down asymmetric toq: Set neqtyp = 0 in the namelist input. Then gato expects to read: Line 1: A date Line 2: A title (like in the g files from EFIT) Line 3: npsi, ntht, neqsym etc. Here, neqsym should be 0 if the equilibrium is asymmetric. Other: ndsksz = 0 (the default) => make one ugta file per matrix block. --------------------------------------------------------------- 3/27/2000 Notes from testing of the workstation version of gato seig compile options for the "fastO5" case: -O5 -arch host -tune host -fast -r8 -i8 Timing benchmarks mesh 1 Cholesky decomp seconds rt1_fastO5 rt1_fastO5 badger_fastO5 ulam_fastO5 cray sv1 ca17n1 ca21n8 ca17n1 ca17n1 60x120 1.75 1.75 2.35 5.6 6.5 80x160 5.75 5.3 6.83 17.1 20.2 100x200 13.9 13.2 15.9 40.6 46.4 120x240 30.0 28.5 33.5 81.7 84.6 130x260 42.3 38.7 44.2 111.7 116.0 140x280 53.9 54.6 59.0 148.3 150x300 70.4 69.3 83.0 193.4 193.0 160x320 93.2 94.2 100.5 248.8 246.0 180x360 153.8 151.3 161 393.3 379.0 200x400 256.1 244.6 256 594.2 640.0 300x600 1421 1432 1587(ca21n8) 2954.4 2750 400x800 4851 5400(ca21n8) 8510 480x960 10465 11600(ca21n8) 16400 550x1100 18257 20563(ca21n8) 29010 620x1240 29853 43313 720x1440 55100 800x1600 83000(est) 168300(ca21n8) 1200x2400 mesh 1 Cholesky decomp (seconds), (mostly case aa37n2, where aa53n5 is used, other mesh densities were found to take the same time as aa37n2) rt1_fastO5 katze_fastO5 badger_fastO5 katze_fastO5 compiler: Linux 1.0 Tru64 5.1 Tru64 5.1 Tru64 5.2 60x120 1.81 1.75 2.33 1.62 80x160 5.48 5.15 6.8 4.8 100x200 13 12 16 11.3 120x240 27.6 24 32 22.8 130x260 41 34 44 31.8 140x280 54 46 59 43.2 150x300 73 60 78 57 160x320 102 77 100 73.8 180x360 154 124 160 118.4 200x400 250 200 247 188 300x600 1441 1568 1555 1533 400x800 4920 5649 5573 (aa53n5) 480x960 10337 12294 12183 (aa53n5) 550x1100 620x1240 720x1440 800x1600 1200x2400 Compaq test drive systems DS20 ES40 badger SV1 katze pcsrt1 500 MHz 667 MHz fastO5 fastO5 fastO5 Tru64 Unix 5.0 Linux Tru64 5.2 Linux 1.0 ca21n8 ca21n8 ha1n5 ha2n5 ha2n5 aa37n5 (isym=0) (isym=0) (isym=0) (isym=0) 60x120 2.09 13.8 80x160 47 100x200 117.5 120x240 29.45 27.5 280 264 130x260 379 140x280 55.4 523 150x300 702 160x320 93.7 112 923 180x360 150.3 1538 200x400 237 2700 2408 300x600 1667 1302 14610 19400 13029 400x800 6497 59000 42218 480x960 12942 104000 88790 550x1100 620x1240 720x1440 800x1600 1200x2400 mesh ugta read rate (MByte/sec) rt1_fastO5 rt1_fastO5 badger_fastO5 ulam_fastO5 katze ca17n1 ca21n8 ca17n1 ca17n1 ha2n5 60x120 259 255 98.7 80x160 233 183 112.6 100x200 232 279 106 120x240 183 87 114.4 130x260 180 122 86.6 140x280 93 263 104.9 150x300 114 49 111.2 160x320 28 88 117.4 180x360 26 21.5 108.3 200x400 21 12.9 21.0 300x600 16 5.07 400x800 15.5 480x960 19.6 550x1100 620x1240 1200x2400 mesh disk space required (Gbyte) gzip compression isym=0 isym = 1 60x120 0.035 0.12 80x160 0.075 0.27 100x200 0.14 0.52 120x240 0.235 reduced to 72% 0.9 130x260 0.297 1.1 140x280 0.367 1.37 150x300 0.448 1.68 160x320 0.539 2.04 180x360 0.76 2.88 200x400 1.033 3.94 300x600 3.40 13.2 400x800 7.94 31.0 480x960 13.62 53.5 550x1100 20.44 620x1240 29.2 720x1440 46(est) 800x1600 63(est) 1200x2400 --------------------------------------------------------------- mapping error messages 1. F should be equal to R*Bt. This error says that the mapping routine is finding a difference above the built in tolerance. ***** Important Warning: Non fatal error in subroutine reseteq at location 32 ***** Error: New sf(nprofl) .ne. rcnt*btor : nprofl = 129 kmap = 0 2. 2 methods for computing q on axis disagree by an amount outside the built in tolerance. Warning: Non fatal error in subroutine setaxs0 at location 2 Error: Estimates for qaxe disagree : kqaxe = 0 nx = 129 2a. This message also means that there was some difficulty computing q on axis. ***** Important Warning: Non fatal error in subroutine psifit at location 3 ***** Error: Iteration limit of 10 exceeded : iter = 10 jp = 401 3.Comparisons of the right and left side of the Grad-Shafranov equation at the grid points for the input equilibrium find some large discrepancies. Global Equilibrium Convergence is 0.440807E-03 for 6104 points in the plasma Number of Poorly Converged Surfaces = 4 Number of Poorly Converged Points = 354 Maximum error at (i,j) = ( 37, 25) = 0.194283E+01 4.Something about the mapping deciding not to look at equilibrium points that are to close to each other in the code that traces the flux surface. I don't really understand the details. Solution: in the gato input file add delpkc = 0.005 ( default is 0.01). Also, the value of ntrymx (default is 1) can be increased to 5 or 10. May need to reduce delpkc still more in some cases. Alan thinks that 0 might work but he doesn't sound exactly certain. Simply reduce it until it works. ***** Important Warning: Non fatal error in subroutine maparc at location 27 ***** Error: Arc points misordered: ktry>max : jval = 300 ktry = 1 ***** Aborting run: Fatal error in subroutine maparc at location 29 ** *** Error: Misordered points in tp : ntry = 1 kmis = 2 5.A polynomial fit to 3 points on the q profile is attempted and fails by some criterion. It switches to a linear fit. This isn't a problem according to Alan Turnbull. ***** Important Warning: Non fatal error in subroutine meshpak at location 9 ***** Error: intqad failed: trying intlin : ker0 = 1 lp = 34 6. These messages are a result of a local dip in q near the edge of the equilibrium. 6a. ***** Important Warning: Non fatal error in subroutine meshpak at location 21 ***** Error: Rational surfaces are misordered : lp = 33 kplp1 = 0 6b. After a polynomial fit to 3 points on the q profile, evaluating q at a location between the 3 points finds a q value below the minimum q value at the 3 points. Warning: Non fatal error in subroutine meshpak at location 37 Error: intqad zq out of range zy1...zy3 : ker1 = 2 nw = 509 7. In the o1gta file this message appears to indicate that the computation of the packing function has not been able to put as many points near each rational surface as it wanted to because a array wasn't dimensioned large enough. This message is ok. The code simply uses fewer points on the packing function and recomputes the packing function again. The number of weight mesh points in swgt nwtot = 1202 has exceeded npts = 1201 At rational surface lp = 59 nwb = 1 nwpt12 = 20 kpst = 4 kpak = 4 8. (Seen in aa34n55) When the equilibrium is cut off at the edge and the pressure profile is spline fitted, a value is negative. The negative value is set to 0. There is a table of pressure values in the o1gta file which can be checked. ***** Important Warning: Non fatal error in subroutine reseteq at location 24 ***** Error: New reset pressure is negative : kpneg = 1 nprofl = 131 9. (Seen in aa34n55) One of the F values (terms in the delta W equation) isn't up/down symmetric. This is tested with an equality test with no accounting for roundoff, so this might not be a severe error. ***** Important Warning: Non fatal error in subroutine symetry at location 34 ***** Error: f7 value not symmetric : kfq07 = 27 jfq07 = 1 10. (Seen when running aa37n5 which is up/down symmetric but setting isym = 0 to make gato run up/down asymmetric.) ***** Important Warning: Non fatal error in subroutine rdtoqmp at location 8 ***** Error: neqsym and isym are inconsistent : neqsym = 1 isym = 0 11.(Seen when running aa37n5 which is up/down symmetric but setting isym = 0 to make gato run up/down asymmetric. Probably symmetry isn't important for this message.) Some quadratic interpolations of psi(q) are having trouble in the packing routines. It reverts to using a linear interpolation. Just means surfaces might not be packed as 'accurately' as they would otherwise. ***** Important Warning: Non fatal error in subroutine meshpak at location 10 ***** Error: s(q) from intqad outside range : ker1 = 2 lp = 27 -------------------------------------------------------------------------- 5/4/2000 If the eigenmode is from the Alven continuum sometimes seigh has trouble converging the inverse iterations. In this case the iteration error value will reach a minimum and then probably slowly increase.This is because of difficulty resulting the narrow spike at the rational surface that is the eigenfunction for the Alven mode.