3D malfunction: allene is not planar

Hi Marvin programmers,

testing now MarvinSketch 5.3.0beta1 I found nearly all old problems with the 3D structures for allenes have been solved :-)

Only for allene itself Clean 3D with hydrogenize produces still the wrong planar structure, see the attached screenshot. Even when starting with a perpendicular structure Clean 3D goes back to the wrong planar structure.

The tool Conformers produces for allene the correct perpendicular structure. So I think the correct algorithm must be present.

For methyl-allene (1.2-butadiene) both Clean 3D with hydrogenize and Conformers produce the correct perpendicular result.

For 1.3-dimethyl-allene (2.3-pentadiene) both procedures produce also the correct structures, and here you get even both enatiomers for this chiral molecule, see second screenshots.

We should have had this new version three weeks ago in a seminar we used MarvinSketch.

Regards, Hans-Ulrich

Dear Hans-Ulrich,

 

Thanks for the quick test. Lot of improvements in the field of 3D coordinate generation will be released in 5.3. Unfortunatelly, the correct handling of allene is not fully implemented: the force field is fixed however the raw coordinate generation still estimates planar geometry around cumulated double bonds. For allene a simple optimization is not able to break this simmetry (actually, this is good).

It is good news that hyperfine or using bigger structures the optimization is able to find the real energy minima.

Regards,

Gabor

Dear Gabor,

I will answer with two posts, here the first:

As shown in the attached image the geometry of allene depends dramatically on the "Hyperfine" option:

Hyperfine Off: wrong planar structure with very high energy.

Hyperfine On: correct perpendicular structure with low energy.

Obviously uses "Clean 3D with hydrogenize" the first method.

Regards, Hans-Ulrich

Dear Gabor,

and now my question:

" ... For allene a simple optimization is not able to break this symmetry (actually, this is good). "

Why is this good ? It cannot be good to have an applet in the internet and many, many users learn "allene is planar", which is absolutely wrong.

Clean 3D with hydrogenize must use better methods, at least for those tiny molecules with high symmetry.

Is it a problem of the symmetry of this molecule ? The procedure does not find the way from the energy hill down to the minimum in the valley ?

Regards, Hans-Ulrich

Dear Hans-Ulrich,

I am sorry if my answer was ambiguous. As i wrote currently planar geometry is estimated around cumulated double bonds during the coordinate generation. This is bug, we will fix it.

The force field itself now handles these situations correctly.

" ... For allene a simple optimization is not able to break this symmetry (actually, this is good). "


Why is this good ?

The fact that the MM optimizaition executed for the estimated planar structure will not break its planarity is the desired behavior.

Is it a problem of the

symmetry of this molecule ? The procedure does not find the way from

the energy hill down to the minimum in the valley ?

The problem is the wrong planar initial geometry generated inside clean3D, not the input itself. The optimization itself can not (and should not) find way down from the top of an energy peak.

As shown in the attached image the geometry of allene depends dramatically on the "Hyperfine" option:

Hyperfine is intended to break out from such situations. Of course it is still incorrect to generate wrong single conformer without using hyperfine.

Clean 3D with hydrogenize must use better methods, at least for those tiny molecules with high symmetry.

I agree that these structures should be handled correctly.

Regards,

Gabor

Dear Gabor,

thank you for the detailed answer. So I can inform my students about the situation.

Regards, Hans-Ulrich

Hi Gabor, having now Marvin 5.4.0.0 I will test old unfixed bugs.

Drawing allene as shown in the attached screen shot and then using Clean 3D Fine with Hydrogenize still procuces a planar structure, which is wrong.

For methyl-allene the correct orthogonal structure is produced as can be seen in the screenshot, too.

I must repeat  a sentence I posted here ten months ago: "It cannot be good to have an applet in the internet and many, many users learn 'allene is planar', which is absolutely wrong."

Please implement a special routine for handling allene.

BTW: The new display with fading the black colour in the background is a hindrance and not a feature. Is it possible to switch it off ? I will test it with larger structures.

Best regards, Hans-Ulrich

Hi, I have an idea here to solve the allene problem.

In earlier posts in this topic I found, the Conformers optimization with Hyperfine = On and Optimzation limit = Very strict produces the correct orthogonal structure for allene.

So a solution may be: Insert in the "Clean 3D Fine with Hydrogenize" code lines with the meaning

if (formula = C3H4) { hyperfineOptimize = true;  optimizationLimit = veryStrict; }

Then for "normal" molecules the normal Clean 3D mechanism works and produces fast good structures and only for this critical tiny molecule it uses this more time consuming methods. When it is doing the same for propyne and cyclopropene, too, this is not disturbing. These are tiny molecules, too.

Regards, Hans-Ulrich

Dear Hans-Ulrich,

 

Thank you for your post.

Hi Gabor, having now Marvin 5.4.0.0 I will test old unfixed bugs. 

I can confirm that we did not fixed this bug yet. Its ticket is kept open however it was not scheduled yet. I agree that this issue should be fixed.

So a solution may be: Insert in the "Clean 3D Fine with Hydrogenize" code lines with the meaning


if (formula = C3H4) { hyperfineOptimize = true;  optimizationLimit = veryStrict; }

 

Unfortunatelly the proposed solution might not fix all structures containing the problematic part (C=C=C) as a fragment .

BTW: The new display with fading the black

colour in the background is a hindrance and not a feature. Is it

possible to switch it off ? I will test it with larger structures.

As far as i know a discussion about this problem is alive in a separate topic.

 

Regards,

Gabor

Dear Gabor,

first a comment to the "fading" background: We found a good compromise in the seperate topic ftopic6994.

And now to the allene probleme. As shown in the first attached image only the three allenes in the upper line persist planar after "Clean 3D with Hydrogenize" . The hydroxy-allene and the mercapto-allene go down to the correct orthogonal structure. The methyl-allene, too.

The second image shows the energies for planar and orthogonal structures of allene itself, calculated using "Tools-Conformers-Without/With Hyperfine". The difference is very high: 188 kJ/mol (50 kcal/mol). The problem is : Clean 3D does not find the way down from the maximum.

Some posts above you had the sentences: "The optimization itself can not (and should not) find way down from the top of an energy peak. " Yes, that's the problem. But: Why "should not" ?

What is the purpose not to go down of an energy peak ?

It seems only a small number of small molecules stays on the top, the other go down and that's correct.

Best regards, Hans-Ulrich

BTW: Tools-Conformers-MMFF64 is not able to find the correct allene structure, too.

Dear Hans-Ulrich,

 

Thanks for the very detailed tests. I believe we discussed the main aspects of this problem in this and other topics:

- Clean3D itself inside produces wrong planar initial geometry for these structures, estimating the (elongated) dihedral angle at a local maximum. This is a bug.

- MM optimization can not break the symmetry of these situations in most cases since gradient is approximately zero. Where non zero gradient is present MM has a chance to find a local minimum.

-Hyperfine is able to divert the symmetric structure away from this extrema, from where MM optimization finds an expected low energy conformer, however this option is and should be switched off by default due to performance reasons.

We will fix the underlying bug, unfortunately it is not scheduled yet.

Regards

Gabor

Dear Gabor, I must comment your sentences, I cannot follow your argumentation.

"Hyperfine is able to divert the symmetric
structure away from this extrema, from where MM optimization finds an
expected low energy conformer, however this option is and should be
switched off by default due to performance reasons"

I think the performance is no problem for seven atoms. Only for these the program is not able to go down from the maximum. So you could implement a single line in your code:

if (number of atoms is less than 8) hyperfine is true;

As can be seen in my last post, only for allenes having 7 atoms your program is not able to go down.

Perhaps it would be better to use a higher number of atoms, 20 or 50. For those tiny molecules the performance is no problem for today computers.

"We will fix the underlying bug, unfortunately it is not scheduled yet."

This means, thousands of users of MarvinSketch, f.e. of your "online tryout", learn the wrong result: allene is planar (No, allene is not planar).

Let me look in the code. I will find the place, where to insert the line above (of course with the variables you are using). It should be done immediately.

BTW: Discussing the rotation barrier diagram, shown in my last post, with a professor of our campus, he said after some "calculations in his mind": The barrier value of 188 kJ/mol seems to be very reasonable !

Best regards, Hans-Ulrich

Dear Hans-Ulrich,

I would like to remind you that our code is not in the public domain. In order to have an insight you need to apply for a position at ChemAxon. We have some open positions .

Seriously speaking, your comments are very highly appreciated, keep doing it, your feedback is very valuable for us. Please note that at present we have no capacity to improve our 3D structure generation. This does not mean that bugs will not be fixed but - as said earlier and elsewhere - related tasks have not yet been scheduled.

Thank your for your understanding and apologies for the inconvenience these issues might cause.

BR

Miklos

Hi Miklos,

having now version 5.4.1.0, there is still the malfunction of MarvinSketch concerning allene. And it's not a "bug". I must cite and modify my sentences given in previous posts:

This means, thousands of users of MarvinSketch, f.e. of your "online tryout", learn a wrong structure for allene.  No, allene is not planar and 1,3-disubstituted allenes are chiral !

So I must change the title of this topic from "bug" to "malfunction".

Regards, Hans-Ulrich

So I must change the title of this topic from "bug" to "malfunction".

Hi Hans-Ulrich,

Indeed, you're right. Please do so.

Regards

Miklos

OK. I have done it to emphasize these problems.

Regards, Hans-Ulrich

Thank you!

Miklos