Mask intrinsic bottom/top contributions when building FK tables#253
Mask intrinsic bottom/top contributions when building FK tables#253Radonirinaunimi wants to merge 2 commits into
Conversation
| with: | ||
| pixi-version: v0.65.0 | ||
| cache: true | ||
| activate-environment: true |
There was a problem hiding this comment.
I'm totally in favour of dropping poetry for pixi.
But I'm against having two competing lock files in the repository. Either one or the other.
There was a problem hiding this comment.
I don't think I agree with this. The point of the pixi config here is to simply and easily provide LHAPDF for the regression in an isolated environment. It doesn't do anything more than that. And as you can see in the CI, poetry is still the one that orchestrates the build/installation.
There was a problem hiding this comment.
So, in this sense, there isn't a competing lock files at all. Of course we might say we want to fully switch to pixi but that'll involve a non-negligible amount of changes that should not be part of this PR.
There was a problem hiding this comment.
Then I would not commit the lockfile to the repo nor I would add a global pixi.toml file.
You can add a pixi subfield to the pyproject.toml if you want to use it for the tests.
| # identity-evolved) at every scale. However, intrinsic bottom and top must never | ||
| # leak into the FK table, so their flavor-basis columns are masked whenever EKO | ||
| # would otherwise treat them as intrinsic (flavor index > nf). Charm is excluded | ||
| # here, since intrinsic charm is the supported feature. |
There was a problem hiding this comment.
I think I disagree with this.
Leaking into the fktable is not a problem if they are hit with a PDF that is 0 at the fitting scale, like would be the case for the bottom.
If this is the way we want to fix it, it should be fixed at the level of yadism (so no bottom in the grid) rather than removing it here.
Either that or the removal should be informed by the theory card.
There was a problem hiding this comment.
In hindsight, I would tend to agree with that for the main reason that doing it in this way will make PineAPPL grids and EKOs (by themselves) completely useless without pineko.
|
I'm not convinced by this PR ...First: does this PR fix the discrepancy we observe? If the answer is no, this PR should be dropped; if the answer is yes, we can accept this PR as a hotfix, but it is NOT the solution - it would rather mask another, deeper problem |
I am not sure what you mean? I explicitly check that this resolve the discrepancy, that's the sole purpose of it.
Again, I am not sure what do you mean. There are various solutions to address the problem, but the question is which approach is more suitable (correct). And because of this #253 (comment), I am now convinced (as @scarlehoff suggest) that this approach NNPDF/yadism#382 is more correct. |
I just wanted to hear that - just wanted to be 100% sure there is a problem with this.
While we know there is "a" problem, I still do not understand what the problem is exactly. we all agree that the right thing is to not account for intrinsic bottom and the best way - in my opinion - is to do so at input scale level, by providing no intrinsic bottom there. I consider that "best" because it is the most general thing and the whole pipeline has been designed around this. From your comparison here, we think that yadism is doing the right thing: if you provide no intrinsic bottom: before=after From the snippet below, we can check that also eko is doing the right thing: if you provide no intrinsic bottom, it remains 0 for all kinematic bins. So: if yadism does what is expected and eko does what is expected, why is this PR a solution? Check eko snippetimport eko
from ekobox.apply import apply_pdf
import pathlib
import lhapdf
pdf = lhapdf.mkPDF("NNPDF40_nnlo_as_01180",0)
with eko.EKO.read(pathlib.Path("./debug/HERA_NC_318GEV_EM_SIGMARED.tar")) as evolution_operator:
evolved_pdfs, _integration_errors = apply_pdf(evolution_operator, pdf)
for ep, op in evolved_pdfs.items():
print(ep)
print(op[5])
import pdb; pdb.set_trace() |
|
@Radonirinaunimi was the problem at the end the fktable optimization putting intrinsic bottom where the light quarks ought to be? |
|
@felixhekhorn I understand your point and I am wondering whether we are being consistent in comparing FK tables computed from different places. To resolve the mystery about the intrinsic bottom PDF, I computed two different grids from When comparing the resulting FK tables, both agree of the order of So I am now convinced that when using exactly the same settings across the board, we (should) get the same results as before and everything is in the end fine. NB: See materials to reproduce the results. |
|
Wait, then there was no bug at all? |
Yes, and all the partial comparisons (partial in the sense that they were not exactly the same settings as before) that we've done so far fooled us in believing there are. |
but what about the comparison done by Juan R. - he was comparing final FK tables, without any intermediate mess and we started from there |
|
Then the fktables in 4.1 were always correct? I'm very confused about these news. I guess the bug was only the original fktable comparison? |
|
And just got the combined FK table from @andrpie for theory ID Check eko snippet❯ pineappl diff 40008005_HERA_NC_318GEV_EM_SIGMARED.pineappl.lz4 4100001803_HERA_NC_318GEV_EM_SIGMARED.pineappl.lz4 NNPDF40_nnlo_as_01180
b x1 x2 x3 O(as^0 a^0)
---+-----+-----+------+------+---------+---------+------------+------------+---------
0 60 60 0.0008 0.0008 0.74111 0.74111 1.4312753e0 1.4304361e0 -5.863e-4
1 90 90 0.0013 0.0013 0.6841 0.6841 1.3934655e0 1.3931751e0 -2.084e-4
2 90 90 0.0015 0.0015 0.59289 0.59289 1.3667532e0 1.3667102e0 -3.146e-5
3 90 90 0.002 0.002 0.44466 0.44466 1.2883208e0 1.2885928e0 2.111e-4
4 120 120 0.0016 0.0016 0.74111 0.74111 1.3639729e0 1.3637052e0 -1.963e-4
5 120 120 0.002 0.002 0.59289 0.59289 1.3245782e0 1.3246726e0 7.129e-5
6 120 120 0.0032 0.0032 0.37055 0.37055 1.1801007e0 1.1805456e0 3.770e-4
7 150 150 0.002 0.002 0.74111 0.74111 1.3273220e0 1.3271648e0 -1.184e-4
8 150 150 0.0032 0.0032 0.46319 0.46319 1.2103763e0 1.2107326e0 2.944e-4
9 150 150 0.005 0.005 0.29644 0.29644 1.0586593e0 1.0591543e0 4.676e-4
10 150 150 0.008 0.008 0.18528 0.18528 9.0466200e-1 9.0517248e-1 5.643e-4
11 150 150 0.013 0.013 0.11402 0.11402 7.6586562e-1 7.6633328e-1 6.106e-4
12 200 200 0.0026 0.0026 0.76011 0.76011 1.2729382e0 1.2728416e0 -7.587e-5
13 200 200 0.0032 0.0032 0.61759 0.61759 1.2319916e0 1.2321576e0 1.348e-4
14 200 200 0.005 0.005 0.39526 0.39526 1.0931042e0 1.0935108e0 3.720e-4
15 200 200 0.008 0.008 0.24704 0.24704 9.3492293e-1 9.3537287e-1 4.813e-4
16 200 200 0.013 0.013 0.15202 0.15202 7.8840992e-1 7.8882570e-1 5.274e-4
17 200 200 0.02 0.02 0.098814 0.098814 6.7773311e-1 6.7809147e-1 5.288e-4
18 200 200 0.032 0.032 0.061759 0.061759 5.7629551e-1 5.7657809e-1 4.903e-4
19 200 200 0.05 0.05 0.039526 0.039526 4.9612075e-1 4.9632664e-1 4.150e-4
20 200 200 0.08 0.08 0.024704 0.024704 4.2757563e-1 4.2770409e-1 3.004e-4
21 200 200 0.13 0.13 0.015202 0.015202 3.6836570e-1 3.6843283e-1 1.823e-4
22 200 200 0.18 0.18 0.010979 0.010979 3.2542633e-1 3.2546681e-1 1.244e-4
23 250 250 0.0033 0.0033 0.74859 0.74859 1.2206804e0 1.2206599e0 -1.681e-5
24 250 250 0.005 0.005 0.49407 0.49407 1.1129491e0 1.1132605e0 2.798e-4
25 250 250 0.008 0.008 0.30879 0.30879 9.5640078e-1 9.5679698e-1 4.143e-4
26 250 250 0.013 0.013 0.19003 0.19003 8.0511372e-1 8.0548853e-1 4.655e-4
27 250 250 0.02 0.02 0.12352 0.12352 6.8973045e-1 6.9005398e-1 4.691e-4
28 250 250 0.032 0.032 0.077199 0.077199 5.8386572e-1 5.8411996e-1 4.354e-4
29 250 250 0.05 0.05 0.049407 0.049407 5.0036874e-1 5.0055325e-1 3.687e-4
30 250 250 0.08 0.08 0.030879 0.030879 4.2914490e-1 4.2925979e-1 2.677e-4
31 250 250 0.13 0.13 0.019003 0.019003 3.6772975e-1 3.6778980e-1 1.633e-4
32 250 250 0.18 0.18 0.013724 0.013724 3.2351729e-1 3.2355343e-1 1.117e-4
33 250 250 0.25 0.25 0.0098814 0.0098814 2.6422942e-1 2.6424615e-1 6.332e-5
34 250 250 0.4 0.4 0.0061759 0.0061759 1.4447912e-1 1.4448129e-1 1.502e-5
35 300 300 0.0039 0.0039 0.76011 0.76011 1.1789454e0 1.1789355e0 -8.406e-6
36 300 300 0.005 0.005 0.59289 0.59289 1.1216257e0 1.1218338e0 1.855e-4
37 300 300 0.008 0.008 0.37055 0.37055 9.7199695e-1 9.7234301e-1 3.560e-4
38 300 300 0.013 0.013 0.22803 0.22803 8.1817250e-1 8.1851321e-1 4.164e-4
39 300 300 0.02 0.02 0.14822 0.14822 6.9925908e-1 6.9955501e-1 4.232e-4
40 300 300 0.032 0.032 0.092638 0.092638 5.8990297e-1 5.9013533e-1 3.939e-4
41 300 300 0.05 0.05 0.059289 0.059289 5.0375749e-1 5.0392575e-1 3.340e-4
42 300 300 0.08 0.08 0.037055 0.037055 4.3039823e-1 4.3050290e-1 2.432e-4
43 300 300 0.13 0.13 0.022803 0.022803 3.6723023e-1 3.6728498e-1 1.491e-4
44 300 300 0.18 0.18 0.016469 0.016469 3.2201008e-1 3.2204299e-1 1.022e-4
45 300 300 0.25 0.25 0.011858 0.011858 2.6194772e-1 2.6196303e-1 5.844e-5
46 300 300 0.4 0.4 0.0074111 0.0074111 1.4205579e-1 1.4205800e-1 1.557e-5
47 400 400 0.0053 0.0053 0.74577 0.74577 1.1018916e0 1.1019293e0 3.420e-5
48 400 400 0.008 0.008 0.49407 0.49407 9.9072547e-1 9.9097588e-1 2.528e-4
49 400 400 0.013 0.013 0.30404 0.30404 8.3738385e-1 8.3766938e-1 3.410e-4
50 400 400 0.02 0.02 0.19763 0.19763 7.1384284e-1 7.1409725e-1 3.564e-4
51 400 400 0.032 0.032 0.12352 0.12352 5.9927399e-1 5.9947465e-1 3.348e-4
52 400 400 0.05 0.05 0.079051 0.079051 5.0906089e-1 5.0920600e-1 2.851e-4
53 400 400 0.08 0.08 0.049407 0.049407 4.3240583e-1 4.3249606e-1 2.087e-4
54 400 400 0.13 0.13 0.030404 0.030404 3.6653658e-1 3.6658386e-1 1.290e-4
55 400 400 0.18 0.18 0.021959 0.021959 3.1976909e-1 3.1979747e-1 8.875e-5
56 400 400 0.25 0.25 0.01581 0.01581 2.5852424e-1 2.5853752e-1 5.137e-5
57 400 400 0.4 0.4 0.0098814 0.0098814 1.3843520e-1 1.3843743e-1 1.606e-5
58 500 500 0.0066 0.0066 0.74859 0.74859 1.0441989e0 1.0442436e0 4.280e-5
59 500 500 0.008 0.008 0.61759 0.61759 9.9682388e-1 9.9697894e-1 1.555e-4
60 500 500 0.013 0.013 0.38005 0.38005 8.5058077e-1 8.5082166e-1 2.832e-4
61 500 500 0.02 0.02 0.24704 0.24704 7.2481115e-1 7.2503566e-1 3.097e-4
62 500 500 0.032 0.032 0.1544 0.1544 6.0655057e-1 6.0672961e-1 2.952e-4
63 500 500 0.05 0.05 0.098814 0.098814 5.1326597e-1 5.1339562e-1 2.526e-4
64 500 500 0.08 0.08 0.061759 0.061759 4.3408700e-1 4.3416767e-1 1.858e-4
65 500 500 0.13 0.13 0.038005 0.038005 3.6614672e-1 3.6618906e-1 1.156e-4
66 500 500 0.18 0.18 0.027448 0.027448 3.1819586e-1 3.1822123e-1 7.974e-5
67 500 500 0.25 0.25 0.019763 0.019763 2.5604808e-1 2.5605999e-1 4.650e-5
68 500 500 0.4 0.4 0.012352 0.012352 1.3580271e-1 1.3580489e-1 1.609e-5
69 500 500 0.65 0.65 0.0076011 0.0076011 2.1307946e-2 2.1301989e-2 -2.796e-4
70 650 650 0.0085 0.0085 0.75564 0.75564 9.7576061e-1 9.7580547e-1 4.597e-5
71 650 650 0.013 0.013 0.49407 0.49407 8.6285482e-1 8.6303639e-1 2.104e-4
72 650 650 0.02 0.02 0.32115 0.32115 7.3733491e-1 7.3752550e-1 2.585e-4
73 650 650 0.032 0.032 0.20072 0.20072 6.1537336e-1 6.1552988e-1 2.543e-4
74 650 650 0.05 0.05 0.12846 0.12846 5.1855454e-1 5.1866859e-1 2.200e-4
75 650 650 0.08 0.08 0.080287 0.080287 4.3638641e-1 4.3645758e-1 1.631e-4
76 650 650 0.13 0.13 0.049407 0.049407 3.6598076e-1 3.6601815e-1 1.022e-4
77 650 650 0.18 0.18 0.035683 0.035683 3.1662226e-1 3.1664457e-1 7.045e-5
78 650 650 0.25 0.25 0.025692 0.025692 2.5339493e-1 2.5340536e-1 4.119e-5
79 650 650 0.4 0.4 0.016057 0.016057 1.3291816e-1 1.3292023e-1 1.553e-5
80 800 800 0.0105 0.0105 0.75287 0.75287 9.2114552e-1 9.2118833e-1 4.647e-5
81 800 800 0.013 0.013 0.60809 0.60809 8.6828521e-1 8.6840441e-1 1.373e-4
82 800 800 0.02 0.02 0.39526 0.39526 7.4688707e-1 7.4704677e-1 2.138e-4
83 800 800 0.032 0.032 0.24704 0.24704 6.2282805e-1 6.2296612e-1 2.217e-4
84 800 800 0.05 0.05 0.1581 0.1581 5.2326046e-1 5.2336233e-1 1.947e-4
85 800 800 0.08 0.08 0.098814 0.098814 4.3864730e-1 4.3871109e-1 1.454e-4
86 800 800 0.13 0.13 0.060809 0.060809 3.6620442e-1 3.6623796e-1 9.158e-5
87 800 800 0.18 0.18 0.043917 0.043917 3.1568270e-1 3.1570258e-1 6.298e-5
88 800 800 0.25 0.25 0.031621 0.031621 2.5155307e-1 2.5156226e-1 3.652e-5
89 800 800 0.4 0.4 0.019763 0.019763 1.3081441e-1 1.3081626e-1 1.418e-5
90 800 800 0.65 0.65 0.012162 0.012162 1.9780981e-2 1.9777045e-2 -1.990e-4
91 1000 1000 0.013 0.013 0.76011 0.76011 8.6677716e-1 8.6681409e-1 4.261e-5
92 1000 1000 0.02 0.02 0.49407 0.49407 7.5641050e-1 7.5653612e-1 1.661e-4
93 1000 1000 0.032 0.032 0.30879 0.30879 6.3161389e-1 6.3173498e-1 1.917e-4
94 1000 1000 0.05 0.05 0.19763 0.19763 5.2916510e-1 5.2925659e-1 1.729e-4
95 1000 1000 0.08 0.08 0.12352 0.12352 4.4176743e-1 4.4182512e-1 1.306e-4
96 1000 1000 0.13 0.13 0.076011 0.076011 3.6696973e-1 3.6699997e-1 8.242e-5
97 1000 1000 0.18 0.18 0.054897 0.054897 3.1510140e-1 3.1511908e-1 5.608e-5
98 1000 1000 0.25 0.25 0.039526 0.039526 2.4990977e-1 2.4991769e-1 3.166e-5
99 1000 1000 0.4 0.4 0.024704 0.024704 1.2875653e-1 1.2875807e-1 1.190e-5
100 1200 1200 0.014 0.014 0.84698 0.84698 8.4504679e-1 8.4505074e-1 4.667e-6
101 1200 1200 0.02 0.02 0.59289 0.59289 7.6294967e-1 7.6305259e-1 1.349e-4
102 1200 1200 0.032 0.032 0.37055 0.37055 6.3963187e-1 6.3974660e-1 1.794e-4
103 1200 1200 0.05 0.05 0.23715 0.23715 5.3495320e-1 5.3504207e-1 1.661e-4
104 1200 1200 0.08 0.08 0.14822 0.14822 4.4509035e-1 4.4514663e-1 1.264e-4
105 1200 1200 0.13 0.13 0.091213 0.091213 3.6816384e-1 3.6819306e-1 7.939e-5
106 1200 1200 0.18 0.18 0.065876 0.065876 3.1507510e-1 3.1509179e-1 5.298e-5
107 1200 1200 0.25 0.25 0.047431 0.047431 2.4890240e-1 2.4890954e-1 2.870e-5
108 1200 1200 0.4 0.4 0.029644 0.029644 1.2725702e-1 1.2725825e-1 9.599e-6
109 1500 1500 0.02 0.02 0.74111 0.74111 7.6786487e-1 7.6792424e-1 7.732e-5
110 1500 1500 0.032 0.032 0.46319 0.46319 6.5088590e-1 6.5098906e-1 1.585e-4
111 1500 1500 0.05 0.05 0.29644 0.29644 5.4382936e-1 5.4391417e-1 1.560e-4
112 1500 1500 0.08 0.08 0.18528 0.18528 4.5057290e-1 4.5062720e-1 1.205e-4
113 1500 1500 0.13 0.13 0.11402 0.11402 3.7062769e-1 3.7065541e-1 7.480e-5
114 1500 1500 0.18 0.18 0.082345 0.082345 3.1580709e-1 3.1582224e-1 4.800e-5
115 1500 1500 0.25 0.25 0.059289 0.059289 2.4821544e-1 2.4822124e-1 2.339e-5
116 1500 1500 0.4 0.4 0.037055 0.037055 1.2568982e-1 1.2569040e-1 4.574e-6
117 1500 1500 0.65 0.65 0.022803 0.022803 1.8123505e-2 1.8121178e-2 -1.284e-4
118 2000 2000 0.0219 0.0219 0.90241 0.90241 7.5042697e-1 7.5038315e-1 -5.840e-5
119 2000 2000 0.032 0.032 0.61759 0.61759 6.6806329e-1 6.6810582e-1 6.366e-5
120 2000 2000 0.05 0.05 0.39526 0.39526 5.5965085e-1 5.5970501e-1 9.677e-5
121 2000 2000 0.08 0.08 0.24704 0.24704 4.6112451e-1 4.6116204e-1 8.139e-5
122 2000 2000 0.13 0.13 0.15202 0.15202 3.7619764e-1 3.7621605e-1 4.893e-5
123 2000 2000 0.18 0.18 0.10979 0.10979 3.1852258e-1 3.1853119e-1 2.703e-5
124 2000 2000 0.25 0.25 0.079051 0.079051 2.4854625e-1 2.4854756e-1 5.289e-6
125 2000 2000 0.4 0.4 0.049407 0.049407 1.2421410e-1 1.2421283e-1 -1.024e-5
126 2000 2000 0.65 0.65 0.030404 0.030404 1.7522198e-2 1.7520224e-2 -1.127e-4
127 3000 3000 0.032 0.032 0.92638 0.92638 6.8977157e-1 6.8984948e-1 1.130e-4
128 3000 3000 0.05 0.05 0.59289 0.59289 5.9315903e-1 5.9326725e-1 1.825e-4
129 3000 3000 0.08 0.08 0.37055 0.37055 4.8621893e-1 4.8629254e-1 1.514e-4
130 3000 3000 0.13 0.13 0.22803 0.22803 3.9107356e-1 3.9110656e-1 8.439e-5
131 3000 3000 0.18 0.18 0.16469 0.16469 3.2745249e-1 3.2746534e-1 3.926e-5
132 3000 3000 0.25 0.25 0.11858 0.11858 2.5240502e-1 2.5240471e-1 -1.252e-6
133 3000 3000 0.4 0.4 0.074111 0.074111 1.2352908e-1 1.2352499e-1 -3.310e-5
134 3000 3000 0.65 0.65 0.045607 0.045607 1.6870296e-2 1.6868429e-2 -1.107e-4
135 5000 5000 0.0547 0.0547 0.90324 0.90324 6.3165220e-1 6.3171677e-1 1.022e-4
136 5000 5000 0.08 0.08 0.61759 0.61759 5.4674633e-1 5.4681614e-1 1.277e-4
137 5000 5000 0.13 0.13 0.38005 0.38005 4.3089912e-1 4.3092363e-1 5.688e-5
138 5000 5000 0.18 0.18 0.27448 0.27448 3.5353543e-1 3.5353285e-1 -7.284e-6
139 5000 5000 0.25 0.25 0.19763 0.19763 2.6663496e-1 2.6661658e-1 -6.894e-5
140 5000 5000 0.4 0.4 0.12352 0.12352 1.2610826e-1 1.2609318e-1 -1.196e-4
141 5000 5000 0.65 0.65 0.076011 0.076011 1.6440344e-2 1.6437607e-2 -1.665e-4
142 8000 8000 0.0875 0.0875 0.90344 0.90344 6.0328574e-1 6.0329545e-1 1.611e-5
143 8000 8000 0.13 0.13 0.60809 0.60809 5.0111483e-1 5.0110331e-1 -2.300e-5
144 8000 8000 0.18 0.18 0.43917 0.43917 4.0243745e-1 4.0239446e-1 -1.068e-4
145 8000 8000 0.25 0.25 0.31621 0.31621 2.9501052e-1 2.9495174e-1 -1.992e-4
146 8000 8000 0.4 0.4 0.19763 0.19763 1.3353184e-1 1.3349434e-1 -2.808e-4
147 8000 8000 0.65 0.65 0.12162 0.12162 1.6513587e-2 1.6508520e-2 -3.069e-4
148 12000 12000 0.13 0.13 0.91213 0.91213 5.7766985e-1 5.7759475e-1 -1.300e-4
149 12000 12000 0.18 0.18 0.65876 0.65876 4.7099736e-1 4.7088760e-1 -2.330e-4
150 12000 12000 0.25 0.25 0.47431 0.47431 3.3766909e-1 3.3754375e-1 -3.712e-4
151 12000 12000 0.4 0.4 0.29644 0.29644 1.4559989e-1 1.4552601e-1 -5.075e-4
152 12000 12000 0.65 0.65 0.18243 0.18243 1.7011056e-2 1.7002012e-2 -5.317e-4
153 20000 20000 0.25 0.25 0.79051 0.79051 4.1550426e-1 4.1523009e-1 -6.599e-4
154 20000 20000 0.4 0.4 0.49407 0.49407 1.7156442e-1 1.7140384e-1 -9.359e-4
155 20000 20000 0.65 0.65 0.30404 0.30404 1.8372179e-2 1.8353664e-2 -1.008e-3
156 30000 30000 0.4 0.4 0.74111 0.74111 2.0046375e-1 2.0019420e-1 -1.345e-3
157 30000 30000 0.65 0.65 0.45607 0.45607 2.0251235e-2 2.0219519e-2 -1.566e-3
158 50000 50000 0.65 0.65 0.76011 0.76011 2.3535306e-2 2.3479050e-2 -2.390e-3 |
|
But this fktable is with this branch, right? |
That'd be my conclusion yes. And the difference of about 2% that we saw between 4.0 and 4.1 are due to quark masses+CKM+(small) Evolution settings. |
No, everything with master (inc. |
3 participants
This PR fixes a bug where FK tables produced for fixed-flavor-number EKOs (e.g. the FONLL
ffns4massivecomponent,NfFF: 4) could leak spurious intrinsic bottom/top contributions into the predictions (see also NNPDF/yadism#380).ekoautomatically treats any heavy-quark flavor with index greater than the operator's activenfas "intrinsic", ie. it assigns it an identity evolution operator (see #L96-L102). This becomes a problem when the convolved grid has nonzero bottom/top-initiated channels.The solution proposed here is to simply mask the corresponding entries of the operator for bottom and top.
TODO: