File size: 3,167 Bytes
cfbe945
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
{
  "smooth_materials": [
    "rotated_stripes",
    "double_inclusion",
    "smooth_checker"
  ],
  "reference_grid": 512,
  "alternative_reference_grid": 384,
  "rescoring_full_grids": [
    32,
    48,
    64,
    96,
    128,
    192,
    256,
    384
  ],
  "error_windows": [
    2.0,
    8.0
  ],
  "timing_time_end": 8.0,
  "samples": 641,
  "smooth_full_grids": [
    32,
    64,
    96,
    128
  ],
  "smooth_reduced_modes": {
    "32": [
      128
    ],
    "64": [
      128
    ],
    "96": [
      128,
      256
    ],
    "128": [
      128,
      256
    ]
  },
  "explicit_full_grids": [
    64,
    96,
    128
  ],
  "explicit_cfl_values": [
    0.25,
    0.125
  ],
  "explicit_rule": "St\u00f6rmer\u2013Verlet for mass-weighted pressure u''=-A u with zero initial derivative. Set dt=output_interval/ceil(output_interval*c_max/(CFL*h)); every output coincides with a time step. The sparse five-point A is applied to all three source columns.",
  "smooth_receiver_tolerances": [
    0.05,
    0.1
  ],
  "interface_full_grids": [
    32,
    64,
    96,
    128,
    256
  ],
  "interface_reduced_grids": [
    32,
    64,
    96,
    128
  ],
  "interface_modes_per_patch": [
    128,
    256
  ],
  "interface_receiver_tolerances": [
    0.01,
    0.05
  ],
  "interface_field_tolerance": 0.05,
  "chunk_duration": 0.5,
  "field_chunk_samples": 16,
  "threads": 2,
  "cpu_affinity": [
    0,
    1
  ],
  "repetitions": 3,
  "norm": "Equal-weight Frobenius norm over all output times, receivers, and source columns; include t=0. Also retain each source error.",
  "timing_policy": "Three sequential independent constructions per candidate after a shared small-case warm-up. Report the median of complete per-repetition totals, not the sum of stage medians. Include setup, basis, projection, propagation, and the requested output. Exclude diagnostic errors, reference evaluation, file I/O, and source generation quadrature defining the interface problem.",
  "smooth_output": "Receiver output only. Reconstructing full fields is not required or charged.",
  "interface_output": "Report receiver-only and receiver-plus-physical-pressure-field costs separately; joint receiver and field tolerances use the latter.",
  "selection": "Candidates and tolerances are fixed. Choose the least expensive tested candidate meeting the error threshold against the stated reference. Report unattained tolerances. The smooth-grid comparison is a finite-reference error.",
  "rational_krylov": "Use exactly the 12 cyclic shifts, damping 0.2, last-three-vector sequence, double reorthogonalization, and source conditioning from accuracy_cost/config.json. Build each requested rank independently.",
  "dimension": "r=n+2*m for both reduced methods; all n^2 pressure degrees of freedom for full methods.",
  "chebyshev_full_grids": [
    32,
    64,
    96,
    128
  ],
  "chebyshev_tail_tolerance": 1e-17,
  "chebyshev_rule": "Even Chebyshev\u2013Bessel polynomial for cos(t sqrt(A)); Gershgorin spectral bound and analytic uniform tail bound, shared verified implementation in propagation/chebyshev_reference.py. Include assembly and all output production."
}