Surface Evolver Documentation

Back to top of Surface Evolver documentation.       Index.

Surface Evolver Debugging

There are several debugging features built into Evolver. Script debugging is meant for ordinary users; the others are meant mostly for Evolver developers (me).

Script Debugging

A 'script' here refers to any Evolver commands, whether they be in procedures or functions.

Debugging may be done the old-fashioned way, by putting print statments in code. Another alternative is to put subcommand in a script to let the user explore the state of the surface at a given point in a script. If you can't or don't want to modify the script, then you can set breakpoints.

Breakpoints

The user may set a breakpoint in an already loaded script with the "set breakpoint" command. The syntax is
  BREAKPOINT scriptname linenumber
where scriptname is the name of the function or procedure and linenumber is the line number in the file where the breakpoint is to be set. There must be executable code on the line, or you will get an error. linenumber may be an expression.

Breakpoints may be unset individually with

  UNSET BREAKPOINT scriptname linenumber
or as a group with
  UNSET BREAKPOINTS 

When a breakpoint is reached, Evolver will enter into a subcommand prompt, at which the user may enter any Evolver commands (although some commands, such as load would be very unwise). To exit from the subcommand prompt, use q or exit or quit.

Single-stepping: At the debug prompt, there is a special single-step command n. which will continue execution to the next line (to be precise, until the line number changes).

Stack trace: At the debug prompt, the whereami command will print a stack trace, showing the sequence of function or procedure calls made to reach the current spot.


Syntax Debugging

The error messages produced by Evolver's command parser try to be helpful, but sometimes are rather cryptic. Frequent errors are misspelled keywords, variables that are spelled the same as keywords, use of uninitialized variables, and missing semicolons. Check these before trying any fancy syntax debugging as described here. It is possible to print out the parsing steps, but this is not for the faint of heart, since it the printout is in very unfriendly YACC format. You will have to have an understanding of how shift-reduce parsing works to have a chance at understanding the output. The toggle debug controls the YACC trace feature. Example:

Enter command: debug
YACC debugging was OFF.
Now ON.
Enter command: print asdf
Starting parse
Entering state 0
Reading a token: Next token is token COMMAND_START_ ()
Shifting token COMMAND_START_ ()
Entering state 2
Reading a token: Next token is token PRINT_ ()
Reducing stack by rule 2 (line 158), -> @1
Stack now 0 2
Entering state 5
Next token is token PRINT_ ()
Shifting token PRINT_ ()
Entering state 180
Reducing stack by rule 454 (line 1555), PRINT_ -> print
Stack now 0 2 5
Entering state 308
Reading a token: Next token is token NEWIDENT_ ()
Shifting token NEWIDENT_ ()
Entering state 11
Reading a token: Now at end of input.
syntax error

Shifting token error ()
Entering state 316
Reducing stack by rule 391 (line 1295), NEWIDENT_ error -> rexpr
ERROR 2381: Syntax error: Unexpected new identifier 'asdf'.
  Input line so far:
print asdf


Memory Debugging

The c command gives a brief printout of memory usage, including element memory as calculated for the number of existing elements, and total allocated data memory. For example, on Windows:

Enter command:  c
Vertices: 50  Edges: 144  Facets: 96  Bodies: 1  Facetedges: 288
Element memory: 40280
Total data memory: 302122 bytes.
If the verbose toggle is on, then more detailed information (which may depend on your system) is printed:

Enter command: verbose
Verbose ON. (was off)
Enter command: c
Vertices: 50  Edges: 144  Facets: 96  Bodies: 1  Facetedges: 288
Element memory: 40280
    vertex size:  152 bytes;  number allocated: 156
      edge size:   80 bytes;  number allocated: 244
     facet size:   96 bytes;  number allocated: 196
      body size:  424 bytes;  number allocated: 154
 facetedge size:   40 bytes;  number allocated: 388
quantity size:    360 bytes
instance size:    656 bytes

blocks in use: 181    memory in use: 360740
blocks free:    16    memory free:    45080
Heap top: 03477000
Heap size: 48.71 MB
Physical memory size: 1072680960 bytes   Virtual memory top: 7FFE0000
Session:       4 blocks,      11200 bytes
Permanent:    77 blocks,     290922 bytes
Temporary:     0 blocks,          0 bytes
Total data memory: 81 blocks, 302122 bytes.

For gurus only. The memdebug command causes verbose information to be printed at every memory allocation, reallocation, or deallocation. Meant for me to use debugging Evolver itself. If Evolver was compiled with MEMSTRINGS defined, then the file and line location of each memory operation is also printed. It also causes heap checking to be done on some systems at each memory operation.


Gradient Debugging

For gurus mostly. The estimate toggle can be used to verify that gradients are being calculated correctly. When on, estimate will, for each 'g' step, print the energy change based on the gradient, that is, calculate the inner product of the velocity and energy gradient at each vertex and sum the products, and then also pring the actual energy change during the motion. For small scales, these numbers should be very close, since energy should change nearly linearly over short scales. For example,
Enter command: estimate
Estimation ON. (was off)
Enter command: m 1e-6
Scale fixed at 1e-006.
Enter command: g 2
Estimated energy change: -9.16981132075472e-006
Actual energy change   : -9.16979039722321e-006
  2. area:  5.99999083020960 energy:  5.99999083020960  scale: 1.00000e-006
Estimated energy change: -9.16972762183562e-006
Actual energy change   : -9.16970669440076e-006
  1. area:  5.99998166050291 energy:  5.99998166050291  scale: 1.00000e-006
However, if volume constraints or quantity constraints are not satisfied, then it may take a few iterations for agreement to be reached:

Enter command: body[1].target := 2
Enter command: estimate
Estimation ON. (was off)
Enter command: m 1e-6
Scale fixed at 1e-006.
Enter command: g 5
Estimated energy change: -1.20000000000000e-005
Actual energy change   :     2.74061897546020
  5. area:  8.74061897546020 energy:  8.74061897546020  scale: 1.00000e-006
Estimated energy change: -8.05951562616423e-006
Actual energy change   : -0.000107080144722715
  4. area:  8.74051189531548 energy:  8.74051189531548  scale: 1.00000e-006
Estimated energy change: -8.05945861736465e-006
Actual energy change   : -8.06006082854083e-006
  3. area:  8.74050383525465 energy:  8.74050383525465  scale: 1.00000e-006
Estimated energy change: -8.05934708554515e-006
Actual energy change   : -8.05931920133673e-006
  2. area:  8.74049577593545 energy:  8.74049577593545  scale: 1.00000e-006
Estimated energy change: -8.05923555500071e-006
Actual energy change   : -8.05920767277257e-006
  1. area:  8.74048771672778 energy:  8.74048771672778  scale: 1.00000e-006
Enter command:

For optimizing scale, when evolution is well behaved, the estimate will be around twice the actual change, since this is the ratio for a perfectly quadratic energy function. Example:

Enter command: estimate
Estimation ON. (was off)
Enter command: g 5
Estimated energy change:    -1.23332147258421
Actual energy change   :   -0.647118913175537
  5. area:  5.35288108682446 energy:  5.35288108682446  scale: 0.233721
Estimated energy change:   -0.420072251955455
Actual energy change   :   -0.203503019926857
  4. area:  5.14937806689761 energy:  5.14937806689761  scale: 0.212928
Estimated energy change:   -0.219347968060792
Actual energy change   :   -0.108607336477709
  3. area:  5.04077073041990 energy:  5.04077073041990  scale: 0.197501
Estimated energy change:   -0.130401658330236
Actual energy change   :  -0.0647701891524211
  2. area:  4.97600054126748 energy:  4.97600054126748  scale: 0.213378
Estimated energy change:  -0.0809651401128813
Actual energy change   :  -0.0401832076359208
  1. area:  4.93581733363156 energy:  4.93581733363156  scale: 0.198322
Enter command:

Beware that non-smooth evolution, for example with one-sided constraints, can invalidate the assumptions behind estimate.


Hessian Debugging

The hessian_quiet toggle can be turned off to print information on memory usage and workload during Hessian factoring ( hessian, hessian_seek, ritz, eigenprobe commands). For example,

Enter command: hessian_quiet off
Enter command: eigenprobe 0
Sparse init alloc: 36897
Expanded hashtable size: 73794.
Expanded hashtable size: 147588.
Sparse entries: 61444  Final hashtable size: 147588
Hash extra probes: 256957
Variables: 12291  Original fill: 61446
Workspace: 8 bytes
Passes through main loop: 34
Total_fill:  538907
Total_flops: 7.25226e+007
Eigencounts:    3 <,  0 ==,  12286 >
There are some differences in output depending on whether the ysmp toggle is on (using the Yale Sparse Matrix Package) or off (using my home-grown minimal degree algorithm).

For gurus only. Evolver has a feature that permits numerical checking of the correctness of the named-quantity Hessian-calculating routines. The hessian_diff toggle causes Hessians to be calculated with finite differences of gradients rather than the routines in question. There are some limitations on what kinds of surfaces the numerical Hessian works on, primarily it won't work with volume or quantity constraints. But since any quantity can be tested as an energy, that does not affect its main purpose.

Some internal variables for debugging and tweaking of my minimal degree algorithm:

mindeg_debug_level
Controls verbosity of debug messages.
mindeg_min_region_size
Smallest size region desired; smaller regions will be merged with parent node. Default 0.
mindeg_margin
How high to go above minimum degree in seeking good elimination. Default 5.

Iteration Debugging

The itdebug command toggles the printing of information during a 'g' command. It prints the scale factor for the move, the progress in convergence of the volume or fixed quantity constraints (in terms of the total error as a multiple of the tolerance), then the energy and volumes resulting from the move. An example using cube.fe with optimizing scale:

Enter command: g
Calculating volgrads.
First move, scale 0.1
Diff: 74.4218    Old_diff: 74.4218
Next diff: 5.17892
Diff: 5.17892    Old_diff: 74.4218
Next diff: 0.350987
first move:
scale1 0.1 energy1     5.31028806222661
Body             target volume        actual volume        pressure
  1                         1      0.999964901275881     2.26415094339623

0th move:
scale1 0 energy1                    6
Body             target volume        actual volume        pressure
  1                         1                      1     2.26415094339623

Doubling scale, scale 0.2
Diff: 284.517    Old_diff: 284.517
Next diff: 38.248
Diff: 38.248    Old_diff: 284.517
Next diff: 4.87311
Diff: 4.87311    Old_diff: 38.248
Next diff: 0.61629
scale2 0.2 energy2     5.11714524527486
Body             target volume        actual volume        pressure
  1                         1      0.999938371022819     2.26415094339623

Doubling scale, scale 0.4
Diff: 1032.7    Old_diff: 1032.7
Next diff: 261.432
Diff: 261.432    Old_diff: 1032.7
Next diff: 59.494
Diff: 59.494    Old_diff: 261.432
Next diff: 13.1576
Diff: 13.1576    Old_diff: 59.494
Next diff: 2.89075
Diff: 2.89075    Old_diff: 13.1576
Next diff: 0.634175
scale2 0.4 energy2     5.89816432999747
Body             target volume        actual volume        pressure
  1                         1       0.99993658248485     2.26415094339623

Final scale: 0.20612
Diff: 301.341    Old_diff: 301.341
Next diff: 41.6649
Diff: 41.6649    Old_diff: 301.341
Next diff: 5.45074
Diff: 5.45074    Old_diff: 41.6649
Next diff: 0.707497
  1. area:  5.12110729535848 energy:  5.12110729535848  scale: 0.206120
Enter command:

Back to top of Surface Evolver documentation.       Index.
Susquehanna University assumes no responsibility for the content of this personal Web page. Please read the disclaimer.