ACOUSTIC COUPLING, APERTURE SIZE AND TWO-SOURCE CONTROL
====================================================

This is a numerical feasibility screen, not a fracture solver or a construction
design. It does not calculate a minimum sticker size capable of splitting a
diamond or boulder. All cases are illustrative; no physical testing occurred.

Run from custom-poc using Python with numpy:

python efficiency/generate_acoustic_screen.py
python -m unittest discover -s efficiency -p test_acoustic_screen.py -v

Outputs: efficiency/results/NUMERICAL_acoustic_screen.json plus actual test log
and report. The reusable functions are in efficiency/acoustic_screen.py.

1. Interface coupling
---------------------
For passive lossless flat halfspaces with scalar longitudinal impedances Z1 and
Z2 and a normally incident plane wave:

transmitted intensity / incident intensity = 4*Z1*Z2/(Z1+Z2)^2
transmitted pressure / incident pressure = 2*Z2/(Z1+Z2)

With assumed near-ambient air Z=413 Rayl and quartz c-axis longitudinal
Z=16.7335 million Rayl, the energy fraction is about 0.00009872, or 0.009872
percent. The transmitted pressure amplitude
is nevertheless almost twice incident pressure. These statements are compatible:
pressure amplitude alone is not energy flux across media of different impedance.

Using nominal water Z=1.5 million Rayl yields about 0.302, or 30.2 percent in this same
interface model. This is not a recommendation to submerge a device or specimen.
The quartz scalar is derived as sqrt(rho*C33) using density 2646.6 kg/m^3 and
C33=105.8 GPa for one natural-quartz specimen in Heyliger et al., Table II,
hosted by NIST. It describes purely elastic longitudinal propagation along the
crystal c-axis; it is not an isotropic elastic-tensor model or fracture
calibration. Other directions require the tensor and appropriate eigenmodes;
real piezoelectric/electrical boundary conditions are also not modeled.
Finite-body resonance, matching layers, incidence
angle, shear conversion, absorption and cracks are absent. These coefficients
are not universal upper bounds for every resonant coupling arrangement.

Sources and provenance:
Quartz density/stiffness: https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=851249
Nominal water impedance: https://www.ippt.pan.pl/repository/open/o2564.pdf
Air-coupled interface/resonance: https://pmc.ncbi.nlm.nih.gov/articles/PMC6567335/
Air 413 Rayl lies within the representative near-ambient 410-440 Rayl range
reviewed in CONTROLLED-SPLIT-FEASIBILITY.txt. Actual atmospheric conditions and
specimen orientation must replace these assumptions for a calibrated model.

No contact-patch efficiency is invented. A bonded/contact device requires its
source impedance, adhesive/contact stiffness, preload, real area, modal loading
and losses to be measured or modeled. Two patches could couple much better than
air while still being constrained by force, displacement, heat and bond stress.

2. A small patch does not automatically make a narrow airborne beam
-----------------------------------------------------------------
The circular-aperture paraxial first-zero radius scale is

r ~= 1.22 * wavelength * range / aperture_diameter.

At 40 kHz in air (343 m/s), wavelength is 8.575 mm. A 20 mm aperture at 1 m gives
a radius scale of 0.523 m. This is already a wide angle, so it is explicitly
flagged outside the paraxial screen; it is not a validated focal-radius result.

Algebraically demanding a 1 cm radius at 1 m gives an aperture of about 1.046 m,
which also falls outside the small-aperture/range geometry. The module therefore
does not present that number as a valid manufacturing dimension. A full wave
calculation with the actual radiator/focusing geometry is needed in either case.
The 0.2 ratios used for the small-angle warning gates are heuristic, not proof
that all wave-optics assumptions hold. Unfocused radiators additionally need a
near-field/Fresnel check. Higher frequency reduces wavelength but changes air
attenuation, transducer construction, and which specimen modes can be driven.

3. What two independently driven sources can control
----------------------------------------------------
At a fixed frequency, define G as a complex response matrix. Each column is the
measured spatial field produced by one source for a known complex input. Rows
are spatial response locations. A chosen source vector u produces G*u in a
linear model. The function linear_field_control fits a desired target using
least squares with optional regularization and an input-norm budget.

With two channels the rank is at most two, so an arbitrary many-location field
usually cannot be specified independently. The example shows both an exactly
representable target and a target outside a synthetic matrix's column space.
It does NOT prove that two sources can never break an object in a chosen way.
A crack is not a linear displacement field; its initiation and evolution depend
on stress, geometry, anisotropy and defects, and can change the response itself.

Input norm is not energy. The arbitrary example budget of 0.2 has no conversion
to watts, joules, physical patch dimensions or safe operating limits. Measuring
G in volts per command volt gives only electrical transfer response until the
force/velocity channels and input power are independently calibrated.

Practical sequence
------------------
Measure each source separately at several receiver positions and frequencies.
Repeat after remounting. Construct G(f) with phase and uncertainty. Compare
candidate positions and target fields using the numerical residual and required
source commands, then check source force, displacement, thermal and bond limits.
Only then add specimen-specific crack initiation/propagation data and validate
the fracture model. This is the path to a defensible minimum size calculation;
the present screen establishes neither a universal minimum nor universal
impossibility.
