Ashley Kalkowski

DESTROYING BOULDERS / DIRECT RESPONSE EDITION M03

Ask the
full system.

Keep the same specimen and frequency grid. Solve the full discrete equations at every point. Then compare the meshes without a retained-mode approximation.

Research direction: Ashley Kalkowski.
Numerical resolution study · no physical tests.
322 current direct points completed · broad-band accuracy unresolved.

Read the direct-grid comparison

SAME STEEL / SAME LOAD / SAME FREQUENCIES

Two meshes. One fixed question.

How different are their directly calculated responses?

n106,000 tetrahedra
3,630 free displacement DOFs
n1210,368 tetrahedra
6,084 free displacement DOFs

161 original frequencies per mesh
20,000–200,000 Hz
322 completed, numerically valid points

FULL DISCRETE SYSTEM ≠ CONTINUUM OR PHYSICAL TRUTH
Spatial, damping and boundary assumptions still apply.

20 mmUnchanged cube side length
1 NPrescribed top-face force · peak
161Original frequency samples per mesh
0Physical or fracture tests

New: M04 targeted spatial refinement. Three meshes test six selected frequencies from this study. Inspect the new comparisons and retained 3D field. This original full-grid M03 edition and its failed criteria remain unchanged.

01 / REMOVE ONE APPROXIMATION

Direct mesh difference.
Across the original grid.

M02 compared spatial resolutions at a fixed 144-mode cutoff. That difference still contained modal truncation. M03 uses direct discrete solves at every original grid frequency to remove that reduction from the comparison.

The full-grid direct comparison fails the 5% screen.

The largest scaled complex-compliance difference is 165.281% between n10 and n12. The same-mesh 144-mode comparisons also fail: 10.014% on n10 and 9.842% on n12. All 322 computations passed their numerical validity checks; that does not make their discretization sufficiently accurate.

These percentages describe scaled differences in complex response. They are not energy efficiency, fracture probability, device performance, strength or a bound on continuous-frequency peaks.

Four panels show direct compliance on two meshes, direct mesh differences, modal versus direct errors, and a complex decomposition. The three whole-grid scalar screens fail the 5 percent threshold.
Actual saved scalar data, all 161 samples per mesh. Lines connect the sampled points; no values between those points were solved for this figure. The normalized differences remain substantial even after the retained-mode reduction is removed. Figure inputs and hashes; executable plot source, SVG and plotted CSV are included in the core archive.
Frozen revision 2 outcomes; each declared screen uses a 5% threshold
QuantityComparison scopeValue and worst sampled frequencyStatus
Direct complex compliancen10 versus n12, all 161 original samples165.281% at 131.759 kHzFAILED
144 modes versus direct / n10Same mesh; complex scalar compliance10.014% at 200.000 kHzFAILED
144 modes versus direct / n12Same mesh; complex scalar compliance9.842% at 200.000 kHzFAILED
Direct compliance at 20 kHzOne predeclared sample; n10 versus n120.147% at 20.000 kHzPASSED
Volume/time RMS stress at 20 kHzOne predeclared sample; n10 versus n120.211% at 20.000 kHzPASSED

A difference between two direct finite-element meshes is a numerical comparison, not a measured error against the real specimen. A maximum on 161 frequencies is not a bound on response peaks between those samples. The two passing 20 kHz checks apply to that one sample, not a contiguous frequency band. No whole-grid stress acceptance criterion was declared.

Read the denominator.

For each mesh, Sn(f) = max(|Hd,n(f)|, 0.01 maxf|Hd,n|). The direct mesh comparison uses S12; each modal/direct comparison uses that mesh’s own direct Sn. This floor prevents near-zero compliance from dominating through an arbitrarily tiny denominator.

A complex identity, not additive shares.

M = D + E10 − E12, where D is the direct mesh difference and E is each mesh’s modal-reduction difference. Panel four uses S12 for every term. Its 165.406% modal-mesh maximum is distinct from M02’s historical 155.515% value, which used the modal n12 denominator. The magnitudes cannot be added as causal percentages.

02 / KEEP THE BOUNDARIES VISIBLE

What is applied.
What remains assumed.

The specimen is the same idealized three-dimensional steel cube. This edition changes the numerical route to its harmonic response; it does not add a fracture law or a working delivery device.

APPLIED MODEL INPUT

A prescribed interface force.

Uniform top-face traction has a 1 N peak resultant. The bottom face is fixed in all three displacement components. The harmonic load alternates sign, so a physical interface would need to transmit the appropriate push and pull.

MATERIAL / LOSS

Unchanged linear steel.

Young’s modulus 210 GPa, density 7,850 kg/m³ and Poisson’s ratio 0.30. The inherited Rayleigh damping is illustrative and unmeasured. No plasticity, damage, hardness, flaws or crack propagation is modeled.

ENERGY BOUNDARY

Only the ideal specimen boundary.

Any computed work or power refers to this prescribed mechanical loading. The electrical input, actuator losses, attachment and acoustic transmission are outside the model. Whole-device efficiency cannot be inferred.

Direct does not mean transient.

The calculation concerns steady harmonic response. It does not simulate switching the drive on, a timed impact sequence, crack formation or energy accumulation during a real startup.

The hardware studies remain context.

E01 assemblies and EL01 wiring and parts remain available. They have not been shown to deliver this 20–200 kHz prescribed force. No new hardware actuation or calibration occurs in M03.

Method background: COMSOL documents selected-eigenmode superposition; Ansys describes full direct harmonic equations and the steady-state scope. These are explanatory references. This project’s results come from its own published implementation, not those commercial solvers.

03 / PRESERVE THE EXECUTION HISTORY

An interrupted run.
A separately documented repair.

A controller file-update error stopped the first attempt. Its retained points remain part of the record, and its missing points are not silently filled from another run.

ATTEMPT 1 / PRESERVED

233 valid points. 89 unrun.

The controller encountered a Windows file-access error while replacing its progress record. Four selected full fields had been saved. The partial review reconstructed those four fields and checked the 233 retained scalar records; it performed zero separate direct reference solves.

Whole-grid accuracy screens were not evaluated. This execution failure does not establish numerical instability or specimen failure.

REVISION 2 / COMPLETED SEPARATELY

File handling repaired. Accuracy failure retained.

The separately frozen revision completed all 322 points, with zero invalid or unrun points and all six inherited anchor replays passing. The specimen, grid, force, scientific equations and accuracy criteria stayed unchanged. The I/O repair added bounded retry handling for mutable progress metadata; immutable result records were retained separately.

The 233 earlier points are repeated conditions in the history, not 233 extra independent frequencies or specimens. Revision 2’s complete results do not fill or alter the original partial attempt.

Public failure documentation removes only private absolute filesystem paths, with the transformation bound to the original record hashes. The private originals are retained. The history package preserves all 233 original scalar reports and four original fields. A pre-execution status-enum documentation gap was clarified; it was not a failed numerical preflight. The final repaired preflight passed 21 tests.

04 / RETAIN THE WHOLE GRID

Every declared point.
Including failures.

The final record distinguishes completed computations, numerical validity and accuracy outcomes. Every original frequency remains visible, including the samples with large discrepancies.

322 / 322 current points are numerically valid. All three whole-grid scalar accuracy screens fail.

Each mesh contributes 161 direct harmonic solutions. Six selected full displacement fields are retained at indices 0, 80 and 160 on both meshes. The other 316 current full fields were computed but not retained; their scalar records remain available. There are zero new eigensolutions, zero time-domain integrations and zero physical tests in this matrix.

The independent review passed 999 checks across all 322 scalar/report records, reconstructs the six retained fields, and separately solves the two midpoint reference cases. It is a separate implementation within this project, sharing NumPy/SciPy; not an outside laboratory or physical validation. These are different coverage levels, not independent re-solves of all 322 points. The review and its public-core replay are linked below.

Inspect all 161 frequency rows (322 current direct solutions)

Both mesh statuses are completed_valid on every displayed row. Values below are rounded for reading; the CSVs retain the original values. H is complex compliance; only its magnitude is shown in the first two response columns. Error columns include phase through complex differences.

Original frequency grid, not selected peaks. Percent differences use the declared direct normalizers.
IndexkHz|Hd10| m/N|Hd12| m/NDirect mesh %n10 modal/direct %n12 modal/direct %
020.0002.4890e-102.4926e-100.1475.3175.657
120.2902.4952e-102.4989e-100.1485.3045.644
220.5842.5016e-102.5053e-100.1485.2915.629
320.8822.5082e-102.5119e-100.1495.2775.614
421.1852.5151e-102.5188e-100.1495.2635.599
521.4922.5222e-102.5260e-100.1505.2485.584
621.8042.5296e-102.5334e-100.1515.2335.568
722.1202.5372e-102.5411e-100.1515.2175.551
822.4402.5452e-102.5490e-100.1525.2015.534
922.7662.5534e-102.5573e-100.1535.1845.516
1023.0962.5619e-102.5659e-100.1545.1675.498
1123.4302.5708e-102.5747e-100.1545.1505.479
1223.7702.5800e-102.5840e-100.1555.1325.460
1324.1152.5895e-102.5935e-100.1565.1135.440
1424.4642.5994e-102.6035e-100.1565.0945.419
1524.8192.6097e-102.6138e-100.1575.0745.398
1625.1792.6204e-102.6245e-100.1575.0535.376
1725.5432.6316e-102.6357e-100.1575.0325.354
1825.9142.6432e-102.6474e-100.1575.0105.331
1926.2892.6553e-102.6595e-100.1574.9885.307
2026.6702.6680e-102.6722e-100.1564.9645.282
2127.0572.6813e-102.6855e-100.1544.9405.256
2227.4492.6956e-102.6997e-100.1524.9145.228
2327.8472.7112e-102.7149e-100.1374.8865.199
2428.2512.7221e-102.7237e-100.2524.8665.183
2528.6602.7327e-102.7400e-100.2734.8485.152
2629.0762.7503e-102.7567e-100.2324.8175.121
2729.4972.7677e-102.7737e-100.2174.7875.090
2829.9252.7854e-102.7913e-100.2114.7575.059
2930.3582.8037e-102.8095e-100.2084.7265.026
3030.7992.8227e-102.8286e-100.2084.6954.993
3131.2452.8426e-102.8485e-100.2084.6624.958
3231.6982.8633e-102.8693e-100.2094.6294.922
3332.1572.8850e-102.8911e-100.2114.5944.886
3432.6232.9077e-102.9139e-100.2134.5594.848
3533.0962.9315e-102.9378e-100.2154.5224.809
3633.5762.9564e-102.9628e-100.2184.4844.769
3734.0632.9825e-102.9891e-100.2214.4464.727
3834.5573.0100e-103.0167e-100.2244.4054.684
3935.0573.0388e-103.0458e-100.2284.3644.640
4035.5663.0692e-103.0764e-100.2324.3214.594
4136.0813.1012e-103.1085e-100.2364.2774.547
4236.6043.1349e-103.1425e-100.2414.2314.499
4337.1353.1705e-103.1783e-100.2464.1844.448
4437.6733.2081e-103.2162e-100.2514.1364.396
4538.2193.2479e-103.2562e-100.2564.0854.343
4638.7733.2899e-103.2986e-100.2624.0344.287
4739.3353.3346e-103.3436e-100.2693.9804.230
4839.9053.3820e-103.3913e-100.2763.9254.171
4940.4843.4323e-103.4421e-100.2833.8674.110
5041.0713.4860e-103.4962e-100.2913.8084.047
5141.6663.5432e-103.5539e-100.3003.7473.982
5242.2703.6044e-103.6155e-100.3103.6843.915
5342.8823.6698e-103.6816e-100.3203.6193.845
5443.5043.7400e-103.7524e-100.3313.5523.773
5544.1353.8154e-103.8286e-100.3433.4823.698
5644.7743.8967e-103.9106e-100.3563.4103.621
5745.4233.9845e-103.9993e-100.3703.3353.541
5846.0824.0796e-104.0953e-100.3853.2583.459
5946.7504.1828e-104.1997e-100.4023.1783.374
6047.4274.2952e-104.3134e-100.4213.0953.285
6148.1154.4182e-104.4378e-100.4423.0103.194
6248.8124.5531e-104.5743e-100.4652.9213.099
6349.5204.7017e-104.7249e-100.4902.8293.001
6450.2384.8662e-104.8916e-100.5192.7342.899
6550.9665.0492e-105.0772e-100.5512.6352.794
6651.7055.2539e-105.2849e-100.5872.5332.684
6752.4545.4844e-105.5190e-100.6282.4272.571
6853.2155.7456e-105.7846e-100.6752.3172.453
6953.9866.0441e-106.0885e-100.7292.2032.331
7054.7686.3884e-106.4393e-100.7922.0852.205
7155.5626.7896e-106.8489e-100.8661.9622.074
7256.3687.2630e-107.3330e-100.9551.8351.937
7357.1857.8299e-107.9138e-101.0621.7021.795
7458.0148.5204e-108.6231e-101.1931.5651.648
7558.8559.3798e-109.5087e-101.3571.4221.495
7659.7081.0478e-091.0645e-091.5701.2731.336
7760.5731.1930e-091.2154e-091.8521.1181.170
7861.4511.3938e-091.4257e-092.2450.9570.998
7962.3421.6894e-091.7383e-092.8270.7900.819
8063.2462.1669e-092.2510e-093.7720.6160.632
8164.1623.0642e-093.2409e-095.5560.4360.439
8265.0925.3163e-095.8709e-0910.0550.2510.243
8366.0361.4745e-081.6122e-0829.1280.0910.088
8466.9937.1102e-096.2699e-0914.6890.1880.227
8567.9643.3846e-093.1581e-097.3210.3950.452
8668.9492.1598e-092.0598e-094.8970.6190.693
8769.9491.5631e-091.5073e-093.7180.8560.947
8870.9631.2113e-091.1758e-093.0231.1051.214
8971.9919.7937e-109.5491e-102.5661.3671.496
9073.0358.1458e-107.9667e-102.2501.6451.793
9174.0936.9004e-106.7925e-101.6531.9422.104
9275.1676.0805e-105.9315e-102.7632.2052.411
9376.2575.2987e-105.1859e-102.1822.5312.758
9477.3624.6715e-104.5832e-101.9282.8733.123
9578.4844.1564e-104.0833e-101.7923.2303.507
9679.6213.7237e-103.6614e-101.7043.6073.913
9780.7763.3544e-103.3002e-101.6434.0064.343
9881.9463.0354e-102.9875e-101.6034.4294.800
9983.1342.7568e-102.7140e-101.5774.8795.286
10084.3392.5114e-102.4727e-101.5645.3595.805
10185.5622.2934e-102.2581e-101.5635.8716.361
10286.8022.0985e-102.0660e-101.5736.4206.956
10388.0601.9229e-101.8928e-101.5937.0107.598
10489.3371.7638e-101.7356e-101.6257.6478.290
10590.6321.6188e-101.5921e-101.6538.3378.931
10691.9451.4854e-101.4618e-101.4699.0918.937
10793.2781.3683e-101.3428e-101.5939.1648.943
10894.6301.2552e-101.2307e-101.5229.1708.949
10996.0021.1500e-101.1267e-101.4489.1768.955
11097.3941.0523e-101.0297e-101.4049.1828.962
11198.8059.6085e-119.3867e-111.3779.1898.969
112100.2378.7480e-118.5285e-111.3639.1968.976
113101.6907.9339e-117.7148e-111.3619.2038.983
114103.1647.1594e-116.9386e-111.3729.2108.991
115104.6606.4179e-116.1933e-111.3969.2178.998
116106.1775.7031e-115.4722e-111.4369.2259.006
117107.7165.0085e-114.7684e-111.4959.2339.015
118109.2774.3268e-114.0740e-111.5779.2419.023
119110.8613.6497e-113.3798e-111.6889.2499.032
120112.4682.9659e-112.6732e-111.8419.2589.041
121114.0992.2578e-111.9395e-112.0509.2679.051
122115.7521.5795e-111.3185e-111.8309.2769.060
123117.4301.1403e-116.3646e-123.4899.2869.070
124119.1324.7426e-127.3412e-124.0029.2969.081
125120.8591.0218e-111.7930e-115.0119.3069.091
126122.6112.1608e-113.2733e-116.9349.3179.102
127124.3883.7162e-115.4613e-1110.8259.3289.114
128126.1916.0567e-119.3150e-1120.3469.3399.126
129128.0211.0320e-101.8676e-1046.7249.3517.888
130129.8762.1210e-104.4134e-1083.7086.5093.343
131131.7594.3577e-102.0692e-10165.2813.1727.140
132133.6691.8781e-101.0887e-1051.1117.3719.177
133135.6061.0212e-106.8586e-1120.9359.4019.191
134137.5726.4864e-114.5141e-1112.2349.4159.205
135139.5664.2471e-112.8250e-118.9239.4299.220
136141.5892.5980e-111.4665e-117.8239.4449.236
137143.6421.2807e-111.1421e-116.7619.4599.252
138145.7241.4869e-112.3081e-117.7499.4759.268
139147.8362.7179e-115.4114e-1117.2479.4919.286
140149.9796.5478e-111.3398e-1043.9319.5089.303
141152.1531.7648e-102.1715e-1087.8327.9596.921
142154.3581.5196e-101.0236e-1037.7479.2619.341
143156.5967.7266e-116.0686e-1110.7219.5629.361
144158.8664.8416e-114.0514e-114.9459.5829.381
145161.1683.2663e-112.7890e-112.9619.6029.402
146163.5052.2068e-111.8725e-112.0819.6239.425
147165.8751.4032e-111.1462e-111.6359.6459.447
148168.2797.5861e-125.6826e-121.3959.6679.471
149170.7183.7276e-124.1820e-121.2729.6909.496
150173.1936.8315e-127.6685e-120.5789.7149.522
151175.7031.1028e-111.2484e-111.5889.7399.548
152178.2501.5476e-111.7840e-111.4929.7659.576
153180.8342.1076e-112.3106e-111.4949.7939.605
154183.4552.6101e-112.9629e-112.2279.8219.635
155186.1143.3048e-113.7171e-112.5599.8509.666
156188.8124.1184e-114.6498e-113.3049.8809.698
157191.5495.1418e-115.9179e-114.8849.9129.732
158194.3266.5630e-117.9077e-118.7719.9459.767
159197.1428.8452e-111.1757e-1022.6639.9799.804
160200.0001.3394e-101.2549e-1062.51210.0149.842

The first attempt remains incomplete and has no whole-grid result. The complete revision does not repair the underlying broad-band accuracy limitations. The original M02 failed accuracy screens and M01 3D field context remain unchanged.

NEXT WEAKNESS / PLANNED, NOT EXECUTED

Spatial accuracy still needs work.

The response remains sensitive to the mesh across this frequency band even after modal reduction is removed. The next edition should freeze a bounded refinement and verification plan around the sensitivity demonstrated here, while testing retained-mode error separately. Full-band predictive accuracy, other-material performance and fracture claims remain withheld. This is proposed follow-up work: M03 adds no further study, changes no numerical source and retunes no acceptance gate.

05 / EXECUTABLE EVIDENCE

Open the source.
Check the comparison.

The packages bind the exact plans, source, scalar records, retained fields, attempt history and independent review. Downloading or reproducing them does not convert a failed accuracy screen into physical validation.