When the IRC calculation doesn’t work…

Share This:

In searching reaction pathways, the most frustrating thing is when IRC calculation does not work well.

opt = ts works well, single imaginary vibration! We can get information such as reaction energy a little more! and so on……

However IRC calculations are always pulling your feet!

Trial and error is necessary when IRC calculation does not go well, but their efforts are not always rewarded. But I summarized what I think is better to try at least.

If there are better solution besides the method described in this post, please write it in the comment field by all means. Then we can share information!

Some people keep secret such error handling methods as researcher or laboratory-specific techniques, but I think such things are only harmful to the progress of science. Let’s expose it more and more.

日本語版IRC 計算がうまくいかない時

Let’s change the stepsize!

Maxstep=30 for intermediate structure optimization, maxstep=20 for IRC calculation and maxstep=10 for transition state structure optimization are considered to be good.

If the IRC calculation does not work, try smaller step size and it would work out sometimes.

Don’t use rcfc

Rcfc that is the keyword to read information from chk file of the transition state optimization, sometimes doesn’t work. In that cace, use calcfc instead of rcfc.

By the way, there is an urban legend that calcall works when calcfc does not work, but it has never worked well for my model so far …  also, calcall is very time-consuming…

People who give you advice to use calcall have never used calcall …… Don’t believe them.

Use the other version of gaussian

It is well known that gaussian09 has a bug for IRC calculation.

So let’s use gaussian03 !
Even for gaussian09, there some difference between g09 B.01 and g09 D.01. So using different revision of gaussian is also a good idea!

We expected that IRC calculation’s algorithm would be improved in gaussian16, but it seems that it wasn’t improved so far …


Let’s say clearly, Gaussian’s IRC is bad.

It is much better to use GRRM’s IRC than trial and error in Gaussian’s IRC. GRRM’s IRC is always working well.

When we calculated super complicated theozyme model, GRRM’s IRC didn’t reach to the intermediate, but it proceeded 19 steps which is enough for further optimization.

Use other algorithms

Several other IRC algorithms are installed in Gaussian. If you’re messed up with the IRC calculation, probably it’s better to use the other algorithms.

See also: Gaussian’s IRC page

In g16, EulerPC is set as default, but LQA and DVV are also available. Note that GRRM traces IRC by means of Page and McIver (References 1 and 2) which is LQA algorithm.

Specifying IOp (2/12 = 3) may also be helpful.

When it just does not work

Let’s compromise when trying the above method and it does not work.

It may be one way to search the reaction path by methods other than QM, such as MD or NEB?

Also, the transition state structure you have got might be slightly different from the true transition state. It is worth considering to search TS again.

See also: ref.3

In nearly all of these cases, attempted IRC calculations only ran a few steps before stopping at spurious minima, behavior which we often experience when working with flat regions of potential energy surfaces.

Even so, for those who can not give up

I think that making PES by comprehensive calculation is one way to convince people. If you could write a script to move atoms along the vector direction of imaginary vibration and opt & freq, you could easily make PES !

Good Luck on your IRC calculation!


  1. “On evaluating the reaction path Hamiltonian” M. Page and J. W. McIver Jr., J. Chem. Phys. 1988, 88, 922-35. DOI: 10.1063/1.454172
  2. “Following steepest descent reaction paths – the use of higher energy derivatives with ab initio electronic-structure methods” M. Page, C. Doubleday Jr., and J. W. McIver Jr. J. Chem. Phys. 1990, 93, 5634-42. DOI: 10.1063/1.459634
  3. “Pentalenene formation mechanisms redux”
    Michael W. Lodewyk, Dan Willenbring and Dean J. Tantillo Org. Biomol. Chem. 2014, 12, 887–894.

Related Articles

  1. Ryzen Threadripper Gaussian16 Benchmark
  2. How many kcal/mol ?
  3. Why atomic units (a.u.) are used in QM calculation?
  4. Building a PC with Ryzen Threadripper 1950x for Scientific Computation

Leave a Reply