Why does the sum of the per-residue total energy not match the whole score?

At the bottom of a scored PDB file is a POSE_ENERGIES_TABLE. The sum of total, which is already a dot product with the weights, does not match the total score.

Here is an example in Python3, either with or without PyRosetta —This is a question about the C++ not the Python bindinds, hence my posting in Rosetta General

## with PyRosetta

import pandas as pd

pose = pyrosetta.pose_from_pdb('1ubq.r.pdb')

pyrosetta.rosetta.core.pose.remove_nonprotein_residues(pose) # there are no ligands

print(pose.num_chains(), pose.num_jump()) # there is 1 chain and 0 jumps.

pose.energies().clear_energies()

scorefxn = pyrosetta.get_fa_scorefxn()

single = scorefxn(pose)

scores = pd.DataFrame(pose.energies().residue_total_energies_array())

summed = sum(scores.total_score)

print(f'sum={summed}, call={single}, from_energy={pose.energies().total_energy()}')

# sum=-265.05077538177227, call=-307.92166266135933, from_energy=-307.92166266135933

pose.dump_scored_pdb('scored.pdb', scorefxn)



## agnostic

import re

from io import StringIO

with open('scored.pdb') as fh:

block = re.search('#BEGIN_POSE_ENERGIES_TABLEs+([wW]+)#END_POSE_ENERGIES_TABLEs+',

fh.read())

.group(1)

totals = pd.read_csv(StringIO(block), sep=' ').total

print(f'expected={totals[1]}, summed={sum(totals[2:])}')

# expected=-307.922, summed=-265.05079000000006

What exactly causes this difference?

There are no non-zero terms absent from the per residue table. The above is a single chain protein with no ligands.

While I cannot figure out where `total_energies_` intercepts with the `residue_total_energies_` attribute, in the ScoreFunction.cc file there is:

pose.energies().total_energy() = pose.energies().total_energies().dot( weights_ );

Which means that residue_total_energies_ and total_energies_ are calculated separately somewhere. And beyond that, I cannot figure out.

Personally, I am interested in this as I was tring to wrap my head around the fact `scorefxn.get_sub_score` in PyRosetta is also non additive (example below) and I believe this lies at the heart of it.

trues = pyrosetta.rosetta.core.select.residue_selector.TrueResidueSelector().apply(pose)

all_sub = scorefxn.get_sub_score(pose, trues)

ResidueSpanSelector = pyrosetta.rosetta.core.select.residue_selector.ResidueSpanSelector

fore = ResidueSpanSelector(1, 38).apply(pose)

aft = ResidueSpanSelector(39, 76).apply(pose)

scorefxn.get_sub_score(pose, fore) + scorefxn.get_sub_score(pose, aft), all_sub

# -118.6, -214.1