Learn how to ask questions correctly

Alex Granovsky

gran@classic.chem.msu.su

Dear Dobromir,

I'm sorry for huge delay with my reply.

You can obtain necessary information examining the output from the last step of CI optimization with conic code.

Here is a sample for CI in allene molecule optimized using conic code:

Norm(a2) = 0.0591802683 Norm(b2) = 0.0041383997 Norm(a1) = 0.0591118891 Norm(b1) = 0.0041244390 Alpha(a2,b2) = 0.9975690170 Alpha(a1,b1) = 0.9940447540 Alpha(a1,a2) = 0.9999372915 Alpha(b1,b2) = 0.9868473636 Alpha(a1,b2) = 0.9982753908 Alpha(a2,b1) = 0.9949129342 Lambda 1 = 0.5749404178 Lambda 2 = 0.5745876057 Lambda sc = 0.5745891827 Lambda = 0.5745891827 DELTA ENERGY IS 0.000099939393

To interpret the output, one needs take in mind that:

- a2 and a1 denote gradients of the first electronic state at current and previous geometries correspondingly.
- Similarly, b2 and b1 denote gradients of the second electronic state at current and previous geometries
- Norm(a2) denotes a norm of vector a2 etc...
- Alpha(a2,b2) denotes a cosine of angle formed by vectors a2 and b2 etc...
- Various Lambdas are approximations to Lagrange multiplier. The best approximation is denoted simply as Lambda.

The most important information one can extract from these numbers are

the norms of gradients at the MECI and the angle between gradients.

From the fragment of the output above, you can see that gradient

of the second state is small and hence we are close to an extremum

on its PES. Hence it is likely that MECI is peaked. However this may

require additional investigation e.g. an attempt of optimization

of a geometry of the second electronic state using the same state-

averaging.

On the other side, consider situation when both gradients are

large and collinear i.e. with Alpha(a2,b2) approaching unity.

This would correspond to a sloped MECI.

> What is the easiest way to find the g and h vectors, since the

> algorithm does not use them ?

The g vector is simply the difference of gradients for two states.

It can be computed using runtyp=gradient with both istate and jstate

set. There will be three gradients in the output, namely (g1 - g2)/2,

g1 and g2 (in that order). Here, g1 is gradient for the first state

and g2 is gradient for the second state.

It is currently impossible to find h vector with Firefly.

>There is a $HESS group in the PUNCH file, at the end of the calculation, can I learn from it ?

No, you cannot learn from it. It is very approximate.

>Also if I try vibrational analysis, would the correct way after state averaged CI run be:

>removing $mcaver $conic, and $track, along with method=conic in $statpt, and keeping the rest of the input the same, while adding the vibration specific input ?

>Should I use $CONTRL NUMDER=.T. $END ?

>That would keep things at nstate=2 istate=1 type of calculation, I guess.

It is not very good idea to perform vibrational analysis at MECI.

It is better to use some point near MECI. You need to remove $conic

group and method=conic keeping the rest intact while adding

the vibration specific input. You need not to use numder.

>Or can I just run a single point with the exact same input as that from the CI search at the found geometry, but also adding hssend=1 at $statpt. The CI search does switch between algorithms at some point, will the previous algorithm being active in a case like this, and change my results significantly ?

No you cannot use hssend=1

Kind regards,

Alex Granovsky

On Thu Apr 16 '15 7:32pm, Dobromir Antonov Kalchevski wrote

-----------------------------------------------------------

>Hello,

>I successfully found a conical intersection with your program. Now I am trying to figure out how to characterize it - is it sloped, peaked or seamed ? What is the easiest way to find the g and h vectors, since the algorithm does not use them ? Is it possible, if the CI is practically at the peak of the ground state surface, yet not completely (let's say 10 degrees deviation from a perfect middle geometry for the ground state TS, on the reaction path coordinate (my TS is product-similar, not reagent-similar, meaning may be 20 degrees in the wrong direction)), to still go forward through the saddle point and enter the other product. Sometimes a peaked CI has more than one TS around it and more than one product, I'm just trying to figure out what is the chance that I'm getting product from the CI.

>There is a $HESS group in the PUNCH file, at the end of the calculation, can I learn from it ?

>Also if I try vibrational analysis, would the correct way after state averaged CI run be:

>removing $mcaver $conic, and $track, along with method=conic in $statpt, and keeping the rest of the input the same, while adding the vibration specific input ?

>Should I use $CONTRL NUMDER=.T. $END ?

>That would keep things at nstate=2 istate=1 type of calculation, I guess.

>Or can I just run a single point with the exact same input as that from the CI search at the found geometry, but also adding hssend=1 at $statpt. The CI search does switch between algorithms at some point, will the previous algorithm being active in a case like this, and change my results significantly ?

>Thank you in advance, and Best regards,

>Dobromir A Kalchevski

Mon Apr 27 '15 1:43am

This message read