xref: /original-bsd/usr.bin/pascal/pdx/test/parall.p (revision a1c2194a)
1(*
2 * Copyright (c) 1980 The Regents of the University of California.
3 * All rights reserved.
4 *
5 * %sccs.include.redist.c%
6 *
7 *	@(#)parall.p	5.1 (Berkeley) 04/16/91
8 *)
9
10program Parall(input,output);
11
12{Declare constants for unit conversions, convergence tests, etc.}
13
14const	SQRT2 = 1.4142136;	{Square root of 2}
15	TWOPI = 6.2831853;	{Two times pi}
16	MINALPHA = 0.001;	{Minimum y-step size}
17	ROUGHLYZERO = 0.001;	{Approximation to zero for convergence}
18	YTHRESHOLD = 40.0;	{Heuristic constant for thresholding}
19	N = 8;			{Vector and matrix dimension}
20
21
22{Declare types for circuit elements, vectors, and matrices.}
23
24type	VSOURCE = record
25			ampl: real;	 {Volts (peak)}
26			freq: real;	 {Radians/second}
27			xindex: integer;	{Index for x value}
28			yindex: integer;	{Index for y value}
29		  end;
30
31	RLPAIR = record
32			r: real;	 {Ohms}
33			l: real;	 {Henries}
34			islope: real;	 {Amps/second}
35			invariant: real; {Trapezoidal invariant}
36			lasttime: real;  {Most recent time}
37			xindex: array [1..2] of integer;  {x value indices}
38			yindex: array [1..2] of integer;  {y value indices}
39		 end;
40
41	CAPACITOR = record
42			c: real;	 {Farads}
43			vslope: real;	 {Volts/second}
44			invariant: real; {Trapezoidal invariant}
45			lasttime: real;  {Most recent time}
46			xindex: array [1..2] of integer;  {x value indices}
47			yindex: array [1..2] of integer;  {y value indices}
48		    end;
49
50	VECTOR = array [1..N] of real;
51
52	MATRIX = array [1..N,1..N] of real;
53
54
55{Declare global variables for central simulation information.}
56
57var	ground: VSOURCE;	{Ground -- a fake source}
58	itcount: integer;	{Main routine iteration count}
59	update: integer;	{Update loop counter for main}
60	optimcount: integer;	{Number of optimization steps}
61	newtoncount: integer;	{Number of Newton steps}
62	v1: VSOURCE;		{Voltage source V1}
63	rl1: RLPAIR;		{R1/L1 resistor-inductor pair}
64	rl2: RLPAIR;		{R2/L2 resistor-inductor pair}
65	c1: CAPACITOR;		{Capacitor C1}
66	a: MATRIX;		{Global matrix A}
67	b: MATRIX;		{Global matrix B}
68	jac: MATRIX;		{Global Jacobian matrix}
69	x: VECTOR;		{Global vector of dependents}
70	xnext: VECTOR;		{Next x-vector for simulation}
71	y: VECTOR;		{Global vector of independents}
72	ynext: VECTOR;		{Next y-vector for simulation}
73	gradient: VECTOR;	{Global gradient vector}
74	h: real;		{Time step value}
75	time: real;		{Current time}
76	lastychange: real;	{YStep's most recent y-change}
77	timestep: integer;	{Current timestep number}
78	maxsteps: integer;	{Number of time steps to run}
79        oldxnorm: real;		{Old one-norm of x-vector}
80	newxnorm: real;		{New one-norm of x-vector}
81	closenough: boolean;   	{Flag to indicate convergence}
82
83
84
85
86{The following procedure initializes everything for the program based
87 on the little test circuit suggested by Sarosh.  The user is asked
88 to specify the simulation and circuit parameters, and then the matrix
89 and vector values are set up.}
90
91procedure InitializeEverything;
92var i,j: integer;
93    rtemp: real;
94begin
95
96{Ready the input and output files (almost nil for Berkeley).}
97writeln(output);
98writeln(output,'*** Simulation Output Record ***');
99writeln(output);
100writeln(output);
101
102{Initialize convergence test/indication variables.}
103oldxnorm := 0.0;
104newxnorm := 0.0;
105lastychange := 0.0;
106
107{Get desired time step size for simulation.}
108readln(input,h);
109writeln(output,'h (Seconds): ',h:12:7);
110
111{Get desired number of time steps for simulation.}
112readln(input,maxsteps);
113writeln(output,'maxsteps: ',maxsteps:4);
114
115{Get parameters for source V1 and initialize the source.}
116with v1 do
117   begin
118   readln(input,rtemp);
119   writeln(output,'V1 (Volts RMS): ',rtemp:9:3);
120   ampl := rtemp * SQRT2;
121   readln(input,rtemp);
122   writeln(output,'f (Hertz): ',rtemp:9:3);
123   freq := rtemp * TWOPI;
124   xindex := 1;
125   yindex := 1;
126   end;
127
128{Get parameters for R1/L1 pair and initialize the pair.}
129with rl1 do
130   begin
131   readln(input,r);
132   writeln(output,'R1 (Ohms): ',r:9:3);
133   readln(input,l);
134   writeln(output,'L1 (Henries): ',l:12:7);
135   islope := 0.0;
136   invariant := 0.0;
137   lasttime := -1.0;        {Negative to force first update}
138   xindex[1] := 2;
139   yindex[1] := 2;
140   xindex[2] := 3;
141   yindex[2] := 3;
142   end;
143
144{Get parameters for R2/L2 pair and initialize the pair.}
145with rl2 do
146   begin
147   readln(input,r);
148   writeln(output,'R2 (Ohms): ',r:9:3);
149   readln(input,l);
150   writeln(output,'L2 (Henries): ',l:12:7);
151   islope := 0.0;
152   invariant := 0.0;
153   lasttime := -1.0;	   {Negative to force first update}
154   xindex[1] := 4;
155   yindex[1] := 4;
156   xindex[2] := 5;
157   yindex[2] := 5;
158   end;
159
160{Get parameters for capacitor C1 and initialize the device.}
161with c1 do
162   begin
163   readln(input,c);
164   writeln(output,'C1 (Farads): ',c:12:7);
165   vslope := 0.0;
166   invariant := 0.0;
167   lasttime := -1.0;	   {Negative to force first update}
168   xindex[1] := 6;
169   yindex[1] := 6;
170   xindex[2] := 7;
171   yindex[2] := 7;
172   end;
173
174{Initialize the ground node.}
175with ground do
176   begin
177   ampl := 0.0;
178   freq := 0.0;
179   xindex := 8;
180   yindex := 8;
181   end;
182
183{Zero out all the vectors and matrices.}
184for i := 1 to N do
185   begin
186   x[i] := 0.0;
187   y[i] := 0.0;
188   for j := 1 to N do
189      begin
190      a[i,j] := 0.0;
191      b[i,j] := 0.0;
192      jac[i,j] := 0.0;
193      end;
194   end;
195
196{Initialize the A matrix.}
197a[1,2] := -1.0;
198a[2,3] := 1.0;
199a[2,4] := -1.0;
200a[3,5] := 1.0;
201a[4,7] := 1.0;
202a[5,1] := 1.0;
203a[7,6] := 1.0;
204a[8,8] := 1.0;
205
206{Initialize the B matrix.}
207b[1,1] := 1.0;
208b[3,7] := -1.0;
209b[4,8] := -1.0;
210b[5,2] := 1.0;
211b[6,3] := 1.0;
212b[6,4] := 1.0;
213b[7,5] := 1.0;
214b[8,6] := 1.0;
215
216{Initialize the Jacobian matrix.}
217rtemp := h / (2.0 * rl1.l  +  rl1.r * h);
218jac[2,2] := rtemp;
219jac[3,3] := rtemp;
220jac[2,3] := -rtemp;
221jac[3,2] := -rtemp;
222rtemp := h / (2.0 * rl2.l  +  rl2.r * h);
223jac[4,4] := rtemp;
224jac[5,5] := rtemp;
225jac[4,5] := -rtemp;
226jac[5,4] := -rtemp;
227jac[6,6] := -1.0;
228jac[7,7] := 1.0;
229jac[7,6] := h / (2.0 * c1.c);
230end;
231
232
233
234
235{The following procedure solves the equation Ax=b for an N x N system
236 of linear equations, where A is the coefficient matrix, b is the
237 right-hand-side vector, and x is the vector of unknowns.  Gaussian
238 elimination with maximal column pivots is used.                     }
239
240procedure Solve(a: MATRIX; b: VECTOR; var x: VECTOR);
241var y,z: real;
242    i,j,k,k1: integer;
243begin
244for k := 1 to N-1 do
245   begin
246   y := abs(a[k,k]);
247   j := k;
248   k1 := k + 1;
249   for i := k1 to N do
250      if abs(a[i,k]) > y then
251         begin
252         j := i;
253         y := abs(a[i,k]);
254         end;
255   for i := k to N do
256      begin
257      y := a[k,i];
258      a[k,i] := a[j,i];
259      a[j,i] := y;
260      end;
261   y := b[k];
262   b[k] := b[j];
263   b[j] := y;
264   z := a[k,k];
265   for i := k1 to N do
266      begin
267      y := a[i,k] / z;
268      a[i,k] := y;
269      for j := k1 to N do a[i,j] := a[i,j]  -  y * a[k,j];
270      b[i] := b[i] - y * b[k];
271      end;
272   end;
273x[N] := b[N] / a[N,N];
274for i := N-1 downto 1 do
275   begin
276   y := b[i];
277   for j := i+1 to N do y := y  -  a[i,j] * x[j];
278   x[i] := y / a[i,i];
279   end;
280end;
281
282
283{The following procedure computes and returns a vector called "deltay",
284 which is the change in the y-vector prescribed by the Newton-Rhapson
285 algorithm.}
286
287procedure NewtonStep(var deltay: VECTOR);
288var phi: VECTOR;
289    m: MATRIX;
290    i,j,k: integer;
291begin
292for i := 1 to N do
293   begin
294   phi[i] := 0.0;
295   for j := 1 to N do
296      begin
297      phi[i] := phi[i]  +  a[i,j] * y[j]  +  b[i,j] * x[j];
298      m[i,j] := -a[i,j];
299      for k := 1 to N do m[i,j] := m[i,j] - b[i,k] * jac[k,j];
300      end;
301
302   end;
303Solve(m,phi,deltay);
304end;
305
306
307
308
309{The following function computes the value of theta, the objective
310 function, given the x and y vectors.}
311
312function ThetaValue(x,y: VECTOR): real;
313var i,j: integer;
314    phielem: real;
315    theta: real;
316begin
317theta := 0.0;
318for i:= 1 to N do
319   begin
320   phielem := 0.0;
321   for j := 1 to N do
322      phielem := phielem  +  a[i,j] * y[j]  +  b[i,j] * x[j];
323   theta := theta  +  phielem * phielem;
324   end;
325ThetaValue := theta;
326end;
327
328
329{The following function computes the theta value associated with a
330 proposed step of size alpha in the direction of the gradient.}
331
332function Theta(alpha: real): real;
333var ythere: VECTOR;
334    i: integer;
335begin
336for i := 1 to N do
337   ythere[i] := y[i]  -  alpha * gradient[i];
338Theta := ThetaValue(x,ythere);
339end;
340
341
342{The following procedure computes the gradient of the objective
343 function (theta) with respect to the vector y.}
344
345procedure ComputeGradient;
346var i,j,k: integer;
347    m: MATRIX;
348    v: VECTOR;
349begin
350{Compute v = Ay + Bx and M = A' + J'B'.}
351for i := 1 to N do
352   begin
353   v[i] := 0.0;
354   for j := 1 to N do
355      begin
356      v[i] := v[i]  +  a[i,j] * y[j]  +  b[i,j] * x[j];
357      m[i,j] := a[j,i];
358      for k := 1 to N do
359         m[i,j] := m[i,j]  +  jac[k,i] * b[j,k];
360      end;
361   end;
362{Compute gradient = 2Mv.}
363for i := 1 to N do
364   begin
365   gradient[i] := 0.0;
366   for j := 1 to N do
367      gradient[i] := gradient[i]  +  m[i,j] * v[j];
368   gradient[i] := 2.0 * gradient[i];
369   end;
370end;
371
372
373{The following procedure computes the bounds on alpha, the step size
374 to take in the gradient direction.  The bounding is done according
375 to an algorithm suggested in S.W.Director's text on circuits.}
376
377procedure GetAlphaBounds(var lower,upper: real);
378var alpha: real;
379    oldtheta,newtheta: real;
380begin
381if ThetaValue(x,y) = 0.0
382   then
383      begin
384      lower := 0;
385
386      upper := 0;
387      end
388   else
389      begin
390      lower := MINALPHA;
391      oldtheta := Theta(lower);
392      upper := MINALPHA * 2.0;
393      newtheta := Theta(upper);
394      if newtheta <= oldtheta then
395         begin
396         alpha := upper;
397         repeat
398            begin
399            oldtheta := newtheta;
400            alpha := alpha * 2.0;
401            newtheta := Theta(alpha);
402            end
403         until (newtheta > oldtheta);
404         lower := alpha / 4.0;
405         upper := alpha;
406         end;
407      end;
408end;
409
410
411{The following function computes the best value of alpha for the step
412 in the gradient direction.  This best value is the value that minimizes
413 theta along the gradient-directed path.}
414
415function BestAlpha(lower,upper: real): real;
416var alphaL,alphaU,alpha1,alpha2: real;
417    thetaL,thetaU,theta1,theta2: real;
418
419begin
420alphaL := lower;
421thetaL := Theta(alphaL);
422alphaU := upper;
423thetaU := Theta(alphaU);
424alpha1 := 0.381966 * alphaU  +  0.618034 * alphaL;
425theta1 := Theta(alpha1);
426alpha2 := 0.618034 * alphaU  +  0.381966 * alphaL;
427theta2 := Theta(alpha2);
428repeat
429   if theta1 < theta2
430      then
431         begin
432         alphaU := alpha2;
433         thetaU := theta2;
434         alpha2 := alpha1;
435         theta2 := theta1;
436         alpha1 := 0.381966 * alphaU  +  0.618034 * alphaL;
437         theta1 := Theta(alpha1);
438         end
439      else
440         begin
441         alphaL := alpha1;
442         thetaL := theta1;
443         alpha1 := alpha2;
444         theta1 := theta2;
445         alpha2 := 0.618034 * alphaU  +  0.381966 * alphaL;
446         theta2 := Theta(alpha2);
447         end
448until abs(thetaU - thetaL) <= ROUGHLYZERO;
449BestAlpha := (alphaL + alphaU) / 2.0;
450end;
451
452
453{The following procedure computes and returns a vector called "deltay",
454 which is the change in the y-vector prescribed by the steepest-descent
455 approach.}
456
457procedure OptimizationStep(var deltay: VECTOR);
458var lower,upper: real;
459    alpha: real;
460    i: integer;
461begin
462ComputeGradient;
463GetAlphaBounds(lower,upper);
464if lower <> upper then
465   begin
466   alpha := BestAlpha(lower,upper);
467   for i:= 1 to N do deltay[i] := - alpha * gradient[i];
468   end;
469end;
470
471
472
473
474{The following function computes the one-norm of a vector argument.
475 The length of the argument vector is assumed to be N.}
476
477function OneNorm(vec: VECTOR): real;
478var sum: real;
479    i: integer;
480begin
481sum := 0;
482for i := 1 to N do  sum := sum + abs(vec[i]);
483OneNorm := sum;
484end;
485
486
487{The following procedure takes a y-step, using the optimization
488approach when far from the solution and the Newton-Rhapson approach
489when fairly close to the solution.}
490
491procedure YStep;
492var deltay: VECTOR;
493    ychange: real;
494    scale: real;
495    i: integer;
496begin
497NewtonStep(deltay);
498ychange := OneNorm(deltay);
499if ychange > YTHRESHOLD
500   then
501{
502      begin
503      OptimizationStep(deltay);
504      ychange := OneNorm(deltay);
505      if ychange > YTHRESHOLD then
506}
507         begin
508         scale := YTHRESHOLD/ychange;
509         for i := 1 to N do deltay[i] := scale * deltay[i];
510	 optimcount := optimcount + 1;
511	 end	 	{;}
512{
513      optimcount := optimcount + 1;
514      end
515}
516   else
517      begin
518      newtoncount := newtoncount + 1;
519      end;
520for i := 1 to N do ynext[i] := y[i] + deltay[i];
521end;
522
523
524
525
526{The following procedure updates the output of a voltage source
527 given the current time.}
528
529procedure VsourceStep(vn: VSOURCE);
530begin
531with vn do
532   xnext[xindex] := ampl * sin(freq * time);
533end;
534
535
536{The following procedure updates the outputs of a resistor-inductor
537 pair given the time step to take...that is, this routine takes a
538 time step for resistor-inductor pairs.  The new outputs are found
539 by applying the trapezoidal rule.}
540
541procedure RLPairStep(var rln: RLPAIR);
542begin
543with rln do
544   begin
545   if (time > lasttime) then
546      begin
547      lasttime := time;
548      invariant := xnext[xindex[1]]  +  (h / 2.0) * islope;
549      end;
550   islope := (y[yindex[1]]  -  y[yindex[2]]  -  r * xnext[xindex[1]]) / l;
551   xnext[xindex[1]] := invariant  +  (h / 2.0) * islope;
552   xnext[xindex[2]] := - xnext[xindex[1]];
553   end;
554end;
555
556
557{The following procedure updates the outputs of a capacitor given the
558 time step...it takes the time step using a trapezoidal rule iteration.}
559
560procedure CapacitorStep(var cn: CAPACITOR);
561var v: real;
562begin
563with cn do
564   begin
565   if (time > lasttime) then
566      begin
567      lasttime := time;
568      v := xnext[xindex[2]]  -  y[yindex[2]];
569      invariant := v  +  (h / 2.0) * vslope;
570      end;
571   vslope := y[yindex[1]] / c;
572   v := invariant  +  (h / 2.0) * vslope;
573   xnext[xindex[1]] := - y[yindex[1]];
574   xnext[xindex[2]] := y[yindex[2]] + v;
575   end;
576end;
577
578
579{The following procedure controls the overall x-step for the
580 specific circuit to be simulated.}
581
582procedure XStep;
583begin
584VsourceStep(v1);
585RLPairStep(rl1);
586RLPairStep(rl2);
587CapacitorStep(c1);
588VsourceStep(ground);
589end;
590
591
592
593
594{The following procedure prints out the values of all the voltages and
595 currents for the circuit at each time step.}
596
597procedure PrintReport;
598begin
599writeln(output);
600writeln(output);
601writeln(output,'REPORT at Time = ',time:8:5,' seconds');
602writeln(output,'Number of iterations used: ',itcount:2);
603writeln(output,'Number of truncations: ',optimcount:2);
604writeln(output,'Number of Newton y-steps: ',newtoncount:2);
605writeln(output,'The voltages and currents are:');
606writeln(output,'  V1 = ',x[1]:12:7,'   I1 = ',y[1]:12:7);
607writeln(output,'  V2 = ',y[2]:12:7,'   I2 = ',x[2]:12:7);
608writeln(output,'  V3 = ',y[3]:12:7,'   I3 = ',x[3]:12:7);
609writeln(output,'  V4 = ',y[4]:12:7,'   I4 = ',x[4]:12:7);
610writeln(output,'  V5 = ',y[5]:12:7,'   I5 = ',x[5]:12:7);
611writeln(output,'  V6 = ',x[7]:12:7,'   I6 = ',y[6]:12:7);
612writeln(output,'  V7 = ',y[7]:12:7,'   I7 = ',x[6]:12:7);
613writeln(output,'  V8 = ',x[8]:12:7,'   I8 = ',y[8]:12:7);
614end;
615
616
617
618
619{Finally, the main routine controls the whole simulation process.}
620
621begin
622InitializeEverything;
623PrintReport;
624for timestep := 1 to maxsteps do
625   begin
626   itcount := 0;
627   optimcount := 0;
628   newtoncount := 0;
629   closenough := FALSE;
630   time := h * timestep;
631   repeat
632      begin
633      itcount := itcount + 1;
634      YStep;
635      XStep;
636      for update := 1 to N do
637	 begin
638	 x[update] := xnext[update];
639	 y[update] := ynext[update];
640	 end;
641      oldxnorm := newxnorm;
642      newxnorm := OneNorm(x);
643      if abs(newxnorm - oldxnorm) <= ROUGHLYZERO
644         then closenough := TRUE;
645      end;
646      if itcount < 4 then closenough := FALSE;
647   until (itcount = 25) or closenough;
648   PrintReport;
649   end;
650end.
651
652