Fix double-counted prey-bin quadrature in getDiet() and getTrophicLevel()

[gustavdelius]

Fixes #474.

The bug

Under second_order_w(params) <- TRUE, setPredKernel() builds ft_pred_kernel_e as the predation kernel integrated over the prey bin, dividing by beta - 1 so that the plain w * dw supplied by the prey vector cancels. mizerEncounter() accordingly keeps the point weight params@w_full * params@dw_full.

getDiet() did not. It weighted its prey vectors with the trapezoidally bin-averaged w, applying the prey-bin quadrature a second time on top of a kernel that already contained it. On a geometric grid bin_average_weight(w) / w is exactly (1 + beta) / 2, so the diet came out uniformly too large by that factor — 9.7% for NS_params. Both branches were affected: the FFT path for the reason above, and the custom-kernel path because there params@pred_kernel is point-sampled while getDiet() weighted it with the bin average.

The default proportion = TRUE hid this, since a uniform factor divides out in the normalisation.

getTrophicLevel() had the same mismatch plus a second one. Its trophic-level-weighted numerator was built from the point-sampled getPredKernel(), while its denominator came from getEncounter(), which uses the bin-integrated kernel once the flag is on. The two are then no longer the same integral, and the trophic levels moved by up to 0.06 when the flag was flipped.

The fix

Both functions now use the plain point weight w * dw, matching mizerEncounter(). The prey-bin integral belongs in the kernel, where setPredKernel() already puts it.

New internal encounter_kernel() returns the kernel that mizerEncounter() actually uses: the point-sampled kernel on the default path, the bin-integrated one under second_order_w, and the stored array when a custom kernel is set. It recovers the bin-integrated weights from ft_pred_kernel_e by inverse FFT rather than duplicating the quadrature code, so it cannot drift from whatever setPredKernel() chose. getTrophicLevel() builds its numerator from that.

getPredKernel() is refactored onto a shared expand_kernel_offsets() helper; its behaviour is unchanged. It remains the point-sampled kernel — the right object for plotting and for supplying a custom kernel — which is why the internal helper is separate.

Effect

Summing getDiet(proportion = FALSE) over prey now reproduces getEncounter() * (1 - getFeedingLevel()) in all four combinations of flag and kernel branch. Trophic levels under the flag now sit ~0.014 below the first-order values, consistently signed, instead of moving by up to ±0.06 with mixed signs.

Nothing changes on the default path.

Tests

Four new tests in test-second_order_summary.R, each run under both settings of the flag and both kernel branches:

• encounter_kernel() contracted with the prey vector reproduces getEncounter()
• getDiet(proportion = FALSE) summed over prey equals the consumption rate
• a predator whose prey all have trophic level 1 comes out at exactly 2

Each fails against the previous code (the trophic-level numerator/denominator ratio was 0.97 rather than 1). Full suite passes: 3545 tests, 0 failures.

🤖 Generated with Claude Code