Weird behavior of LOWER_WALLS #1091

riccardocapelli · 2024-06-28T09:41:15Z

Today I used a LOWER_WALLS potential on a distance and set the exponent of the function to 1.

Looking at the bias, I got a value of 0 when d>threshold, but a negative value (i.e. an attractive potential) when d<threshold.

Is this the expected behavior (and I should put a - in front of the KAPPA) or, as the name of LOWER_WALLS suggests, is this a bug?

Thanks a lot!

Riccardo

Iximiel · 2024-09-09T14:56:40Z

I think the problem is here:

plumed2/src/bias/Walls.cpp

Line 97 in 669793f

 readInputLine( getShortcutLabel() + "_pow_" + argn + ": CUSTOM PERIODIC=NO FUNC=step(-x)*x^" + exp[i] + " ARG=" + getShortcutLabel() + "_scale_" + argn ); 

LOWER_WALLS now gets shorcut to a custom func step(-x)*x^pow so that if pow is odd x^pow<0 when x<0.

whereas UPPER_WALLS to step(x)*x^pow, but UPPER_WALLS does not care about the sign of x in the exponential part

The solution should be to change the custom function to either:
-step(-x)*abs(x)^pow

step(-x)*(-x)^pow.

We need to test the faster between the two and PR it I think (and maybe add a test for this)

GiovanniBussi · 2024-09-09T15:00:39Z

I agree that this is a bug, should be fixed as @Iximiel suggested (the two approaches should have the same measurable performance)

Iximiel · 2024-09-13T07:39:28Z

I think @riccardocapelli just kicked a beehive 🐛

#1109 may not be enough:

I'll use a "fancy nomenclature":

UPPER_WALL is "above", so it should "push down" when you get above it
LOWER_WALL is "below", so it should "push up" when you get under it

With @Puncist we started to set up the most simple test for this (walls on the x coordinate), and we discovered a few interesting things, already in 2.9:

UPPER_WALLS works as intended
LOWER_WALLS pushes "down" instead of "up"

Some examples in spoiler

We are running with this config

type=driver
arg="--plumed plumed.dat --ixyz trajectory.xyz  --debug-forces forces --dump-forces-fmt=%8.4f"


function plumed_regtest_before() {
    for x in {-5..5}; do
        echo "1" 
        echo "100 100 100"
        echo "C $x 0 0"
    done > trajectory.xyz    
}

function plumed_regtest_after() {   
    awk 'NR==1{print} /^ 0 /{print}' forces > forces_0_x
}

forces_0_x are the values of the forces on the x component of atom 1 (I called it 0_x because I think in C...)

And this plumed.dat:

p: POSITION ATOM=1 NOPBC

UPPER_WALLS ARG=p.x AT=0 KAPPA=1.0 EXP=__FILL__ LABEL=uwall
LOWER_WALLS ARG=p.x AT=0 KAPPA=1.0 EXP=__FILL__ LABEL=lwall

PRINT ARG=p.x,*.bias,*.force2 FILE=COLVAR

With these two files, you can reproduce the same thing on your PC.

I checked out the branch v2.9, but on master you can get the same results

EXP=1

forces_0_x

#! FIELDS parameter analytical numerical
 0  -1.0000  -1.0000
 0  -1.0000  -1.0000
 0  -1.0000  -1.0000
 0  -1.0000  -1.0000
 0  -1.0000  -1.0000
 0   0.0000  -1.0000
 0  -1.0000  -1.0000
 0  -1.0000  -1.0000
 0  -1.0000  -1.0000
 0  -1.0000  -1.0000
 0  -1.0000  -1.0000

Also, the numerical derivative is -1 in x=0

COLVAR

#! FIELDS time p.x uwall.bias lwall.bias uwall.force2 lwall.force2
 0.000000 -5.000000 0.000000 -5.000000 0.000000 1.000000
 1.000000 -4.000000 0.000000 -4.000000 0.000000 1.000000
 2.000000 -3.000000 0.000000 -3.000000 0.000000 1.000000
 3.000000 -2.000000 0.000000 -2.000000 0.000000 1.000000
 4.000000 -1.000000 0.000000 -1.000000 0.000000 1.000000
 5.000000 0.000000 0.000000 0.000000 0.000000 0.000000
 6.000000 1.000000 1.000000 0.000000 1.000000 0.000000
 7.000000 2.000000 2.000000 0.000000 1.000000 0.000000
 8.000000 3.000000 3.000000 0.000000 1.000000 0.000000
 9.000000 4.000000 4.000000 0.000000 1.000000 0.000000
 10.000000 5.000000 5.000000 0.000000 1.000000 0.000000

EXP=2

forces_0_x

#! FIELDS parameter analytical numerical
 0  10.0000  10.0000
 0   8.0000   8.0000
 0   6.0000   6.0000
 0   4.0000   4.0000
 0   2.0000   2.0000
 0   0.0000  -0.0000
 0  -2.0000  -2.0000
 0  -4.0000  -4.0000
 0  -6.0000  -6.0000
 0  -8.0000  -8.0000
 0 -10.0000 -10.0000

COLVAR

#! FIELDS time p.x uwall.bias lwall.bias uwall.force2 lwall.force2
 0.000000 -5.000000 0.000000 25.000000 0.000000 100.000000
 1.000000 -4.000000 0.000000 16.000000 0.000000 64.000000
 2.000000 -3.000000 0.000000 9.000000 0.000000 36.000000
 3.000000 -2.000000 0.000000 4.000000 0.000000 16.000000
 4.000000 -1.000000 0.000000 1.000000 0.000000 4.000000
 5.000000 0.000000 0.000000 0.000000 0.000000 0.000000
 6.000000 1.000000 1.000000 0.000000 4.000000 0.000000
 7.000000 2.000000 4.000000 0.000000 16.000000 0.000000
 8.000000 3.000000 9.000000 0.000000 36.000000 0.000000
 9.000000 4.000000 16.000000 0.000000 64.000000 0.000000
 10.000000 5.000000 25.000000 0.000000 100.000000 0.000000

EXP=3

forces_0_x

#! FIELDS parameter analytical numerical
 0 -75.0000 -75.0000
 0 -48.0000 -48.0000
 0 -27.0000 -27.0000
 0 -12.0000 -12.0000
 0  -3.0000  -3.0000
 0   0.0000  -0.0000
 0  -3.0000  -3.0000
 0 -12.0000 -12.0000
 0 -27.0000 -27.0000
 0 -48.0000 -48.0000
 0 -75.0000 -75.0000

COLVAR

#! FIELDS time p.x uwall.bias lwall.bias uwall.force2 lwall.force2
 0.000000 -5.000000 0.000000 -125.000000 0.000000 5625.000000
 1.000000 -4.000000 0.000000 -64.000000 0.000000 2304.000000
 2.000000 -3.000000 0.000000 -27.000000 0.000000 729.000000
 3.000000 -2.000000 0.000000 -8.000000 0.000000 144.000000
 4.000000 -1.000000 0.000000 -1.000000 0.000000 9.000000
 5.000000 0.000000 0.000000 0.000000 0.000000 0.000000
 6.000000 1.000000 1.000000 0.000000 9.000000 0.000000
 7.000000 2.000000 8.000000 0.000000 144.000000 0.000000
 8.000000 3.000000 27.000000 0.000000 729.000000 0.000000
 9.000000 4.000000 64.000000 0.000000 2304.000000 0.000000
 10.000000 5.000000 125.000000 0.000000 5625.000000 0.000000

EXP=10

forces_0_x

#! FIELDS parameter analytical numerical
 0 19531250.0000 19531249.7500
 0 2621440.0000 2621439.9531
 0 196830.0000 196829.9956
 0 5120.0000 5119.9998
 0  10.0000  10.0000
 0   0.0000  -0.0000
 0 -10.0000 -10.0000
 0 -5120.0000 -5120.0002
 0 -196830.0000 -196830.0044
 0 -2621440.0000 -2621440.0469
 0 -19531250.0000 -19531250.2500

COLVAR

#! FIELDS time p.x uwall.bias lwall.bias uwall.force2 lwall.force2
 0.000000 -5.000000 0.000000 9765625.000000 0.000000 381469726562500.000000
 1.000000 -4.000000 0.000000 1048576.000000 0.000000 6871947673600.000000
 2.000000 -3.000000 0.000000 59049.000000 0.000000 38742048900.000000
 3.000000 -2.000000 0.000000 1024.000000 0.000000 26214400.000000
 4.000000 -1.000000 0.000000 1.000000 0.000000 100.000000
 5.000000 0.000000 0.000000 0.000000 0.000000 0.000000
 6.000000 1.000000 1.000000 0.000000 100.000000 0.000000
 7.000000 2.000000 1024.000000 0.000000 26214400.000000 0.000000
 8.000000 3.000000 59049.000000 0.000000 38742048900.000000 0.000000
 9.000000 4.000000 1048576.000000 0.000000 6871947673600.000000 0.000000
 10.000000 5.000000 9765625.000000 0.000000 381469726562500.000000 0.000000

EXP=11

forces_0_x

 0 -107421875.0000 -107421873.5000
 0 -11534336.0000 -11534335.7812
 0 -649539.0000 -649538.9844
 0 -11264.0000 -11263.9996
 0 -11.0000 -11.0000
 0   0.0000  -0.0000
 0 -11.0000 -11.0000
 0 -11264.0000 -11264.0004
 0 -649539.0000 -649539.0156
 0 -11534336.0000 -11534336.1875
 0 -107421875.0000 -107421876.5000

COLVAR

#! FIELDS time p.x uwall.bias lwall.bias uwall.force2 lwall.force2
 0.000000 -5.000000 0.000000 -48828125.000000 0.000000 11539459228515624.000000
 1.000000 -4.000000 0.000000 -4194304.000000 0.000000 133040906960896.000000
 2.000000 -3.000000 0.000000 -177147.000000 0.000000 421900912521.000000
 3.000000 -2.000000 0.000000 -2048.000000 0.000000 126877696.000000
 4.000000 -1.000000 0.000000 -1.000000 0.000000 121.000000
 5.000000 0.000000 0.000000 0.000000 0.000000 0.000000
 6.000000 1.000000 1.000000 0.000000 121.000000 0.000000
 7.000000 2.000000 2048.000000 0.000000 126877696.000000 0.000000
 8.000000 3.000000 177147.000000 0.000000 421900912521.000000 0.000000
 9.000000 4.000000 4194304.000000 0.000000 133040906960896.000000 0.000000
 10.000000 5.000000 48828125.000000 0.000000 11539459228515624.000000 0.000000

and so on

Co-authored-by: Daniele Rapetti <[email protected]>

Iximiel added a commit to Iximiel/plumed2 that referenced this issue Sep 9, 2024

this should solve plumed#1091

c614fd2

GiovanniBussi added the bug label Sep 9, 2024

Iximiel mentioned this issue Sep 9, 2024

FIX - Changing the sign of the argument of the exponential in LOWER_WALLS #1109

Merged

7 tasks

Iximiel mentioned this issue Sep 16, 2024

FIX-2.9 Now the forces from LWALLS should be correct #1111

Merged

7 tasks

Iximiel added a commit to Iximiel/plumed2 that referenced this issue Oct 10, 2024

this should solve plumed#1091

3bd5d92

carlocamilloni pushed a commit that referenced this issue Oct 10, 2024

this should solve #1091 (#1109)

4df397b

Co-authored-by: Daniele Rapetti <[email protected]>

carlocamilloni closed this as completed Oct 10, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Weird behavior of LOWER_WALLS #1091

Weird behavior of LOWER_WALLS #1091

riccardocapelli commented Jun 28, 2024

Iximiel commented Sep 9, 2024

GiovanniBussi commented Sep 9, 2024

Iximiel commented Sep 13, 2024 •

edited

Loading

EXP=1

EXP=2

EXP=3

EXP=10

EXP=11

Weird behavior of LOWER_WALLS #1091

Weird behavior of LOWER_WALLS #1091

Comments

riccardocapelli commented Jun 28, 2024

Iximiel commented Sep 9, 2024

GiovanniBussi commented Sep 9, 2024

Iximiel commented Sep 13, 2024 • edited Loading

EXP=1

EXP=2

EXP=3

EXP=10

EXP=11

Iximiel commented Sep 13, 2024 •

edited

Loading