Geometry optimisation

Basic usage

Add Opt to the keyword line. The example below optimises formaldehyde (HCHO) at B3LYP/SV(P):

! B3LYP D4 def2-SVP Opt

* int 0 1
  C 0 0 0   0.000     0.000     0.00
  O 1 0 0   1.203     0.000     0.00
  H 1 2 0   1.107   122.016     0.00
  H 1 2 3   1.107   122.016   180.00
*

Pure GGA functionals (e.g. BP86) enable the RI approximation automatically, so no extra option is needed. For hybrids, pair the run with RIJCOSX def2/J:

# RI + BP86 — extremely fast optimisation
! BP86 D4 def2-SVP Opt

# Hybrids pair with RIJCOSX
! B3LYP D4 def2-TZVP RIJCOSX def2/J Opt

Coordinate systems — Cartesian vs redundant internal

In ORCA 6 the Opt keyword defaults to redundant internal coordinates. This coordinate system represents the geometry as a chemically meaningful set of bond lengths, angles, and dihedrals, and usually converges fastest.

You can force Cartesian coordinates instead, but only in special cases:

Reading the convergence criteria

Each optimisation step evaluates five convergence indicators. The geometry is declared converged once they all fall below their thresholds.

            *********************HURRAY********************
            ***        THE OPTIMIZATION HAS CONVERGED     ***
            *************************************************

------------------------|Geometry convergence|---------------------------
Item                value                   Tolerance       Converged
---------------------------------------------------------------------------
Energy change       -0.0000001234 Eh        0.0000050000      YES
RMS gradient         0.0000234567 Eh/bohr   0.0001000000      YES
MAX gradient         0.0000456789 Eh/bohr   0.0003000000      YES
RMS step             0.0001234567 bohr      0.0020000000      YES
MAX step             0.0002345678 bohr      0.0040000000      YES
---------------------------------------------------------------------------

Thresholds tighten in three steps: NormalOpt (default), TightOpt, VeryTightOpt.

Constrained optimisation

Use this when you want to freeze part of a reaction path, hold specific bonds or angles fixed, etc. Write the constraints inside a %geom Constraints block:

! B3LYP D4 def2-SVP Opt

%geom Constraints
   { B 0 1 1.25 C }       # freeze 0–1 bond length at 1.25 Å
   { A 2 0 3 120.0 C }   # freeze 2-0-3 angle at 120°
   { D 3 1 0 2 180.0 C } # freeze 3-1-0-2 dihedral at 180°
   { C 5 C }              # freeze atom 5 (full Cartesian)
end
end

* xyzfile 0 1 mol.xyz

The first character inside the braces selects the constraint type:

Wildcards and ranges

Multiple coordinates can be tied down in one go:

%geom Constraints
   { B 3 * C }        # all bonds that involve atom 3
   { B * * C }        # every bond
   { A * 5 * C }      # all angles centred on atom 5
   { C 10:17 C }       # Cartesian-freeze atoms 10 through 17
end
end

Inverse constraints

To release only a few coordinates and fix everything else, list the coordinates you want to be free and add invertConstraints true:

%geom Constraints
   { B 0 1 C }
end
   invertConstraints true   # optimise only the C–O distance, freeze everything else
end

Hydrogen-only / partial optimisation

Useful for large active sites or X-ray structures where you want to relax only the hydrogens:

%geom optimizehydrogens true
end

Relaxed surface scan

Step one coordinate from one value to another and at each point relax everything else — the standard tool for exploring a reaction coordinate or a conformational change.

# Scan the H–O–O–H dihedral of H2O2 from 0° to 180° in 19 points
! B3LYP D4 def2-SVP Opt

%geom Scan
   D 0 1 2 3 = 0.0, 180.0, 19   # dihedral 0 → 180, 19 points
end
end

* xyz 0 1
  H   0.000   0.872   0.873
  O   0.000   0.625  -0.085
  O   0.000  -0.625  -0.085
  H   0.000  -0.872   0.873
*

After the scan ORCA writes the (coordinate value, energy) pairs to basename.relaxscanact.dat, and every intermediate geometry is appended to basename_trj.xyz for visualisation.

Multidimensional scans

Scanning two coordinates simultaneously produces a surface:

%geom Scan
   B 0 1 = 3.0, 1.0, 15
   B 1 2 = 1.0, 3.0, 15
end
end

By default the two scans combine as a grid, giving 15 × 15 = 225 points. To run them synchronously instead (pair the same indices, 15 points total), add Simul_Scan true.

Choice of initial Hessian

The optimisation algorithm needs an approximate Hessian (matrix of second derivatives) at each step. The default is a model Hessian (Almloef), which works well for most organic molecules. For tricky systems other options are worth trying:

%geom
   InHess Almloef      # default, robust
end

Reading a previously computed Hessian

Computing a Hessian cheaply first (e.g. a semi-empirical method) and using it as the initial guess can make a difficult optimisation tractable:

# Step 1: quick Hessian with XTB
! XTB2 NumFreq
* xyzfile 0 1 mol.xyz

$new_job

# Step 2: high-quality optimisation seeded with the XTB Hessian
! B3LYP D4 def2-TZVP Opt TightSCF
%geom
   InHess     Read
   InHessName "mol_job1.hess"
end
* xyzfile 0 1

When the optimisation refuses to converge

Things to try when the optimisation diverges or oscillates:

• Raise the iteration cap: %geom MaxIter 500 end.
• Shrink the maximum step: %geom MaxStep 0.1 end (bohr).
• Switch to Cartesian: try ! COpt instead.
• Improve the initial Hessian: the InHess Read pattern shown above.
• Sanity-check the input geometry: look for unreasonable bond lengths or overlapping atoms.

You now have a stationary point. Whether it really is a minimum — no imaginary frequencies — is what a frequency calculation tells you next.