When run() is given a partial start/stop range, the entire results.bond_autocorrelation curve — not just the x=0 point — comes out uniformly rescaled by n_frames_processed / n_frames_total_trajectory, rather than correctly normalized by the number of frames actually processed.
Reproduction (30-atom backbone, 50,000-frame trajectory), plotting the measured C(x) and fitted curve for two ranges on the same chain:
import MDAnalysis as mda
from MDAnalysis.analysis import polymer
import matplotlib.pyplot as plt
u = mda.Universe("conf.psf", "system.xtc")
bb = u.select_atoms("segid RA0 and name P")
for start, stop in [(1, 25000), (1, 50000)]:
plen = polymer.PersistenceLength([bb])
plen.run(start=start, stop=stop, step=1)
plt.figure()
plt.plot(plen.results.x, plen.results.bond_autocorrelation, "ro", label="Result")
plt.plot(plen.results.x, plen.results.fit, "b-", label="Fit")
plt.title(f"start={start}, stop={stop}")
plt.xlabel("x"); plt.ylabel("C(x)"); plt.legend()
plt.savefig(f"lp_{start}_{stop}.png")
Comparing the two resulting curves point-by-point, every value in the stop=25000 curve is almost exactly half the corresponding value in the stop=50000 curve (e.g. C(0)=0.50 vs 1.00, C(x≈6)=0.27 vs 0.55, C(x≈12)=0.18 vs 0.37) — matching 25000/50000 = 0.5 exactly. This shows the mis-normalization is a constant multiplicative rescaling of the whole array, not a single-point artifact, and it corrupts the resulting exponential fit and reported lp for any partial-range analysis. Only start=0/stop=full trajectory length happens to self-correct (numerator == denominator). No warning or error is raised. MDAnalysis version: 2.10.0
Do you have any misunderstanding?
When run() is given a partial start/stop range, the entire results.bond_autocorrelation curve — not just the x=0 point — comes out uniformly rescaled by n_frames_processed / n_frames_total_trajectory, rather than correctly normalized by the number of frames actually processed.
Reproduction (30-atom backbone, 50,000-frame trajectory), plotting the measured C(x) and fitted curve for two ranges on the same chain:
Comparing the two resulting curves point-by-point, every value in the stop=25000 curve is almost exactly half the corresponding value in the stop=50000 curve (e.g. C(0)=0.50 vs 1.00, C(x≈6)=0.27 vs 0.55, C(x≈12)=0.18 vs 0.37) — matching 25000/50000 = 0.5 exactly. This shows the mis-normalization is a constant multiplicative rescaling of the whole array, not a single-point artifact, and it corrupts the resulting exponential fit and reported lp for any partial-range analysis. Only start=0/stop=full trajectory length happens to self-correct (numerator == denominator). No warning or error is raised. MDAnalysis version: 2.10.0
Do you have any misunderstanding?