![Page 1: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/1.jpg)
Strategies in geometry optimization of solids
Claudio M. Zicovich-WilsonFacultad de Ciencias, U. A. del Edo. de Morelos, Mexico
Ab initio Simulation of Crystalline Systems ASCS2006
September 17-22, 2006 - Spokane, Washington (USA)
![Page 2: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/2.jpg)
1
Geometries
![Page 3: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/3.jpg)
1
Geometries
• The same chemical compound (same atoms, same bonds) may have manyconfigurations in what concerns its nuclear positions.
![Page 4: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/4.jpg)
1
Geometries
• The same chemical compound (same atoms, same bonds) may have manyconfigurations in what concerns its nuclear positions.
• In general these nuclear configurations have different energies, and experi-mentally they appear as an statistical distribution of nuclear arrangements.
![Page 5: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/5.jpg)
1
Geometries
• The same chemical compound (same atoms, same bonds) may have manyconfigurations in what concerns its nuclear positions.
• In general these nuclear configurations have different energies, and experi-mentally they appear as an statistical distribution of nuclear arrangements.
• What of such configurations (geometries) can be adopted as a suitablerepresentative of the compound in order to compute the energy dependentproperties?
![Page 6: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/6.jpg)
1
Geometries
• The same chemical compound (same atoms, same bonds) may have manyconfigurations in what concerns its nuclear positions.
• In general these nuclear configurations have different energies, and experi-mentally they appear as an statistical distribution of nuclear arrangements.
• What of such configurations (geometries) can be adopted as a suitablerepresentative of the compound in order to compute the energy dependentproperties?
⇓
GEOMETRYOPTIMIZATION
![Page 7: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/7.jpg)
2
Potential Energy of a nuclear configuration
According to the Born-Oppenheimer approximation, the static total energy dependsuniquely on the nuclear positions:
E = F (x1, x2, · · · , x3N)
(xi Cartesian coordinates of the N nuclei)
Invariance undertranslation-rotations
−→ Internal degrees of freedom
![Page 8: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/8.jpg)
3
Internal degrees of freedom
Molecules:
M = 3N︸︷︷︸cartesian
−translational︷︸︸︷
3 −rotational︷ ︸︸ ︷3[or 2]
![Page 9: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/9.jpg)
3
Internal degrees of freedom
Molecules:
M = 3N︸︷︷︸cartesian
−translational︷︸︸︷
3 −rotational︷ ︸︸ ︷3[or 2]
3D Crystals:
M = 3N︸︷︷︸cartesian
−translational︷︸︸︷
3 + ( 9 −rotational︷︸︸︷
3 )︸ ︷︷ ︸lattice
![Page 10: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/10.jpg)
4
Coordinate systems
![Page 11: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/11.jpg)
4
Coordinate systems
• There are infinite choices to define the M internal coordinates
![Page 12: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/12.jpg)
4
Coordinate systems
• There are infinite choices to define the M internal coordinates
• The best one should be that who makes easier the study of thepotential energy function
![Page 13: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/13.jpg)
4
Coordinate systems
• There are infinite choices to define the M internal coordinates
• The best one should be that who makes easier the study of thepotential energy function
• Let’s assume that v = (v1, v2, · · · , vM) is the nuclear configurationvector in a given coordinate choice
![Page 14: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/14.jpg)
5
A Potential Energy Hypersurface (PEH)
E = F (v1, v2, · · · , vM)
v0
v0" v0’
t1t2
How to describe the features ofthe PEH?
![Page 15: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/15.jpg)
5
A Potential Energy Hypersurface (PEH)
E = F (v1, v2, · · · , vM)
v0
v0" v0’
t1t2
How to describe the features ofthe PEH?A relevant information (inva-riant under the coordinate sys-tem choice) is provided by thecritical points (v0, v′
0 y v′′0 )
![Page 16: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/16.jpg)
6
Critical Points
Condition:
g(v0) ≡ gi(v0) =∂F
∂vi
(v0) = 0
If v is M -dimensional, there are M + 1 different types of critical points. The mostrelevant in quantum chemistry are:
![Page 17: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/17.jpg)
6
Critical Points
Condition:
g(v0) ≡ gi(v0) =∂F
∂vi
(v0) = 0
If v is M -dimensional, there are M + 1 different types of critical points. The mostrelevant in quantum chemistry are:
• minima (v0, v′0)
![Page 18: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/18.jpg)
6
Critical Points
Condition:
g(v0) ≡ gi(v0) =∂F
∂vi
(v0) = 0
If v is M -dimensional, there are M + 1 different types of critical points. The mostrelevant in quantum chemistry are:
• minima (v0, v′0)
• saddle points (v′′0 )
![Page 19: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/19.jpg)
7
The Hessian Matrix
Hij =∂2F
∂vi∂vj
∣∣∣∣v0
Eigenvalue equation:Htµ = hµtµ
hµ −→ real (surface curvature in the c. p.).tµ are special directions in the coordinate space.
![Page 20: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/20.jpg)
8
The critical point order
The order of a critical point v0 is the number of negative eigenvalues (n) of theHessian in that point.
![Page 21: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/21.jpg)
8
The critical point order
The order of a critical point v0 is the number of negative eigenvalues (n) of theHessian in that point.
• n = 0 −→ minimum
![Page 22: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/22.jpg)
8
The critical point order
The order of a critical point v0 is the number of negative eigenvalues (n) of theHessian in that point.
• n = 0 −→ minimum
• n = 1 −→ saddle point
![Page 23: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/23.jpg)
8
The critical point order
The order of a critical point v0 is the number of negative eigenvalues (n) of theHessian in that point.
• n = 0 −→ minimum
• n = 1 −→ saddle point
• . . .
• n = M −→ maximum
![Page 24: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/24.jpg)
9
Reaction path in a bidimensional PEH
A − B · · · C → A · · · C − B
![Page 25: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/25.jpg)
9
Reaction path in a bidimensional PEH
A − B · · · C → A · · · C − B
rCB
rAB
1
T2
A-B...C
A...B-C
![Page 26: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/26.jpg)
10
The chemical meaning of critical points
Critical points
![Page 27: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/27.jpg)
10
The chemical meaning of critical points
Critical points
minima: Clasic→ 0 K most stable↓ temp.≈ most probable
![Page 28: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/28.jpg)
10
The chemical meaning of critical points
Critical points
minima: Clasic→ 0 K most stable↓ temp.≈ most probable
Quantum→ zero-point correctionthermochemistry
![Page 29: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/29.jpg)
10
The chemical meaning of critical points
Critical points
minima: Clasic→ 0 K most stable↓ temp.≈ most probable
Quantum→ zero-point correctionthermochemistry
saddle point: semi-clasic→ Transition state≈ Activation energyKineticsReaction Path
![Page 30: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/30.jpg)
11
Optimization Methods
![Page 31: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/31.jpg)
11
Optimization Methods
• non-gradient
![Page 32: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/32.jpg)
11
Optimization Methods
• non-gradient
• gradient
![Page 33: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/33.jpg)
12
Non gradient methods: line optimization
1 2 3 4 50
5
10
15
sample pointsquadratic function
![Page 34: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/34.jpg)
13
Extension to multiple parameters
Direction set method
![Page 35: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/35.jpg)
13
Extension to multiple parameters
Direction set method
• Each parameter is separately line-optimized in a given sequence
![Page 36: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/36.jpg)
13
Extension to multiple parameters
Direction set method
• Each parameter is separately line-optimized in a given sequence
• A cycle finishes when all parameters have been line-optimized
![Page 37: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/37.jpg)
13
Extension to multiple parameters
Direction set method
• Each parameter is separately line-optimized in a given sequence
• A cycle finishes when all parameters have been line-optimized
• Convergence is reached when the total energy becomes stable under a givennumerical criterion
![Page 38: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/38.jpg)
14
Convergence of the direction set method in a bidimensionalsurface.
![Page 39: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/39.jpg)
14
Convergence of the direction set method in a bidimensionalsurface.
−→ Directions are not fully independent (conjugate)
![Page 40: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/40.jpg)
15
The concept of conjugate directions.
Second order Taylor decomposition of F :
F (u) = F (0) +∑
i∂F∂ui
ui + 12
∑ij
∂2F∂ui∂uj
uiuj + · · ·≈ c − b · u + 1
2u · H · u
![Page 41: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/41.jpg)
15
The concept of conjugate directions.
Second order Taylor decomposition of F :
F (u) = F (0) +∑
i∂F∂ui
ui + 12
∑ij
∂2F∂ui∂uj
uiuj + · · ·≈ c − b · u + 1
2u · H · u
where
∗ u = v − v0, step
∗ c = F (0) = F (v0), energy at v0
∗ b = −g(v0), gradient at v0
∗ H, Hessian matrix
![Page 42: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/42.jpg)
16
The concept of conjugate directions.
• Approximate gradient:g(v) ≡ g(u) = Hu − b.
![Page 43: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/43.jpg)
16
The concept of conjugate directions.
• Approximate gradient:g(v) ≡ g(u) = Hu − b.
• The function is optimized along direction t. −→ gradient change along displace-ment: δt = εt
δg = H · δt = ε (H · t) .
![Page 44: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/44.jpg)
16
The concept of conjugate directions.
• Approximate gradient:g(v) ≡ g(u) = Hu − b.
• The function is optimized along direction t. −→ gradient change along displace-ment: δt = εt
δg = H · δt = ε (H · t) .
• Previously, the function has been optimized along direction s (the gradient hasbecomed perpendicular to s).
• To keep that condition it is required that g ⊥ s along all the displacement −→
0 = s · δg = sHt.
![Page 45: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/45.jpg)
17
Conjugate directions
• Hessian behaves as a metric (quadratic functions):
〈u |v〉 = uHv
![Page 46: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/46.jpg)
17
Conjugate directions
• Hessian behaves as a metric (quadratic functions):
〈u |v〉 = uHv
• Degree of “dependency” between parameters.
⇓
![Page 47: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/47.jpg)
17
Conjugate directions
• Hessian behaves as a metric (quadratic functions):
〈u |v〉 = uHv
• Degree of “dependency” between parameters.
⇓
scalar product: 〈u |v〉 =
0 fully independent
(conjugate)
1 fully dependent
![Page 48: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/48.jpg)
18
Powell’s quadratically convergent method
![Page 49: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/49.jpg)
18
Powell’s quadratically convergent method
• Scheme similar to direction set method
![Page 50: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/50.jpg)
18
Powell’s quadratically convergent method
• Scheme similar to direction set method
• Directions are modified along the proce-dure to be conjugate (within quadraticbehavior of the function).
![Page 51: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/51.jpg)
18
Powell’s quadratically convergent method
• Scheme similar to direction set method
• Directions are modified along the proce-dure to be conjugate (within quadraticbehavior of the function).
↓ Dependent on the coordinate system choice
![Page 52: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/52.jpg)
19
Gradient methods
![Page 53: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/53.jpg)
19
Gradient methods
• Gradient contains information on what is the direction of the steepest energy descent
![Page 54: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/54.jpg)
19
Gradient methods
• Gradient contains information on what is the direction of the steepest energy descent
• This permits to improve the minimum search
![Page 55: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/55.jpg)
19
Gradient methods
• Gradient contains information on what is the direction of the steepest energy descent
• This permits to improve the minimum search
• The numerical calculation of the gradient is costly and a gain in efficiency with respect
to non gradient methods is not ensured.
![Page 56: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/56.jpg)
19
Gradient methods
• Gradient contains information on what is the direction of the steepest energy descent
• This permits to improve the minimum search
• The numerical calculation of the gradient is costly and a gain in efficiency with respect
to non gradient methods is not ensured.
• In some ab initio methods, the gradient can be obtained through analytic formulae
![Page 57: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/57.jpg)
19
Gradient methods
• Gradient contains information on what is the direction of the steepest energy descent
• This permits to improve the minimum search
• The numerical calculation of the gradient is costly and a gain in efficiency with respect
to non gradient methods is not ensured.
• In some ab initio methods, the gradient can be obtained through analytic formulae
• For periodic systems, the implementation of the analytic gradient is not a mere extension
to the molecular case, as infinite sums appear that are to be numerically approximated
by means of suitable techniques.
![Page 58: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/58.jpg)
19
Gradient methods
• Gradient contains information on what is the direction of the steepest energy descent
• This permits to improve the minimum search
• The numerical calculation of the gradient is costly and a gain in efficiency with respect
to non gradient methods is not ensured.
• In some ab initio methods, the gradient can be obtained through analytic formulae
• For periodic systems, the implementation of the analytic gradient is not a mere extension
to the molecular case, as infinite sums appear that are to be numerically approximated
by means of suitable techniques.
• Crystal2006 includes the implementation of the analytic energy derivatives with respect
to
? Atomic positions: K. Doll, V.R. Saunders, N.M. Harrison, Int. J. Quant. Chem. 82 1
(2001)
? Cell parameters: K. Doll, R. Dovesi, R. Orlando. Theor. Chem. Acc. 112, 394-402
(2004)
![Page 59: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/59.jpg)
20
Steepest descent method
![Page 60: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/60.jpg)
20
Steepest descent method
1. How could one exploit the information contained in the gradient?
![Page 61: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/61.jpg)
20
Steepest descent method
1. How could one exploit the information contained in the gradient?
2. The most straightforward strategy would be the Steepest descent method
![Page 62: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/62.jpg)
20
Steepest descent method
1. How could one exploit the information contained in the gradient?
2. The most straightforward strategy would be the Steepest descent method
3. One takes the opposite direction to the gradient
![Page 63: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/63.jpg)
20
Steepest descent method
1. How could one exploit the information contained in the gradient?
2. The most straightforward strategy would be the Steepest descent method
3. One takes the opposite direction to the gradient
4. A line minimization along such a direction is then performed
![Page 64: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/64.jpg)
20
Steepest descent method
1. How could one exploit the information contained in the gradient?
2. The most straightforward strategy would be the Steepest descent method
3. One takes the opposite direction to the gradient
4. A line minimization along such a direction is then performed
5. Iterate since a given convergence criterion (on Energy or gradient norm) is attained
![Page 65: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/65.jpg)
20
Steepest descent method
1. How could one exploit the information contained in the gradient?
2. The most straightforward strategy would be the Steepest descent method
3. One takes the opposite direction to the gradient
4. A line minimization along such a direction is then performed
5. Iterate since a given convergence criterion (on Energy or gradient norm) is attained
Independent on the coordinate system choice
![Page 66: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/66.jpg)
21
Convergence of the steepest descent method in a bidimensionalsurface.
−→Newton-Raphson and other methods.
![Page 67: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/67.jpg)
22
Quadratic behavior
Points v0 −→ v.
![Page 68: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/68.jpg)
22
Quadratic behavior
Points v0 −→ v.Within the second order approximation of the energy, the gradient in v reads
g(v) = H× (v − v0) + g(v0)
![Page 69: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/69.jpg)
22
Quadratic behavior
Points v0 −→ v.Within the second order approximation of the energy, the gradient in v reads
g(v) = H× (v − v0) + g(v0)
Being v the minimum ⇒ g(v) = 0
(v − v0) = −H−1g(v0)
![Page 70: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/70.jpg)
22
Quadratic behavior
Points v0 −→ v.Within the second order approximation of the energy, the gradient in v reads
g(v) = H× (v − v0) + g(v0)
Being v the minimum ⇒ g(v) = 0
(v − v0) = −H−1g(v0)
Within the quadraticapproximation the minimum canbe reached in ONE step, known
the gradient vector and theHessian matrix
![Page 71: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/71.jpg)
23
Newton-Raphson method
The minimum search is made in steps i, in which the following equation is solved:
vi+1 − vi = −H−1i g(vi)
![Page 72: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/72.jpg)
23
Newton-Raphson method
The minimum search is made in steps i, in which the following equation is solved:
vi+1 − vi = −H−1i g(vi)
↑ A few cycles are required to reach the minimumeven in non quadratic PEHs
![Page 73: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/73.jpg)
23
Newton-Raphson method
The minimum search is made in steps i, in which the following equation is solved:
vi+1 − vi = −H−1i g(vi)
↑ A few cycles are required to reach the minimumeven in non quadratic PEHs
↓ The hessian calculation is costly from the compu-tational point of view.
![Page 74: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/74.jpg)
24
Newton-step control
In non-quadratic functions the Newton step may not be the best choice and shouldbe controlled:
![Page 75: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/75.jpg)
24
Newton-step control
In non-quadratic functions the Newton step may not be the best choice and shouldbe controlled:
• level-shift trust region: A trust radius τ of an hypersphere into which thefunction is supposed quadratically behaved. A parameter µ is computed sothat the displacement vi+1 − vi = −(Hi − Iµ)−1g(vi) is kept within thetrust region.
![Page 76: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/76.jpg)
24
Newton-step control
In non-quadratic functions the Newton step may not be the best choice and shouldbe controlled:
• level-shift trust region: A trust radius τ of an hypersphere into which thefunction is supposed quadratically behaved. A parameter µ is computed sothat the displacement vi+1 − vi = −(Hi − Iµ)−1g(vi) is kept within thetrust region.
• Line search: A parameter αi is found so that a minimum is found alongthe Newton step direction:
vi+1 − vi = −αiH−1i g(vi)
![Page 77: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/77.jpg)
25
Improvements to gradient methods
Along the optimization the changes in the gradient between steps provide usefulinformation on the surface curvature (second derivatives):
![Page 78: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/78.jpg)
25
Improvements to gradient methods
Along the optimization the changes in the gradient between steps provide usefulinformation on the surface curvature (second derivatives):
∗ Conjugate gradient: the information is accumulated implicitly(Polak-Ribiere, Berny –implemented in Crystal–)
![Page 79: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/79.jpg)
25
Improvements to gradient methods
Along the optimization the changes in the gradient between steps provide usefulinformation on the surface curvature (second derivatives):
∗ Conjugate gradient: the information is accumulated implicitly(Polak-Ribiere, Berny –implemented in Crystal–)
∗ Variable metric: an approximation to the inverse of the Hessianmatrix H−1 is built during the optimization process.
![Page 80: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/80.jpg)
25
Improvements to gradient methods
Along the optimization the changes in the gradient between steps provide usefulinformation on the surface curvature (second derivatives):
∗ Conjugate gradient: the information is accumulated implicitly(Polak-Ribiere, Berny –implemented in Crystal–)
∗ Variable metric: an approximation to the inverse of the Hessianmatrix H−1 is built during the optimization process. Severalupdating formulae:
• Fletcher-Powell• Murtagh-Sargent (SR1)• Powell-Symetric-Broyden (PSB)• Broyden-Fletcher-Godfarb-Shanno (BFGS)
![Page 81: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/81.jpg)
26
A quite good Hessian to start the procedure may drasticallyreduce the optimization process
![Page 82: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/82.jpg)
26
A quite good Hessian to start the procedure may drasticallyreduce the optimization process
↓ Identity matrix: positive definite, no structure: very inefficient.
![Page 83: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/83.jpg)
26
A quite good Hessian to start the procedure may drasticallyreduce the optimization process
↓ Identity matrix: positive definite, no structure: very inefficient.
→ Numerical Hessian: close to exact, fast convergence, high cal-culation cost (good for very difficult cases: Transition State –seelater–)
![Page 84: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/84.jpg)
26
A quite good Hessian to start the procedure may drasticallyreduce the optimization process
↓ Identity matrix: positive definite, no structure: very inefficient.
→ Numerical Hessian: close to exact, fast convergence, high cal-culation cost (good for very difficult cases: Transition State –seelater–)
↑ Model Hessian: good and cheap approximation to the actualHessian. Based on valence forcefields. Significant improvementwith respect to Identity matrix.
![Page 85: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/85.jpg)
27
Transition state (TS) optimization
![Page 86: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/86.jpg)
27
Transition state (TS) optimization
•
TS −→ critical point −→ g(v) = 0
↓(v − v0) = −H−1g(v0)
![Page 87: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/87.jpg)
27
Transition state (TS) optimization
•
TS −→ critical point −→ g(v) = 0
↓(v − v0) = −H−1g(v0)
• Hessian is not positive definite −→ search is not alwaysdownwards
![Page 88: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/88.jpg)
27
Transition state (TS) optimization
•
TS −→ critical point −→ g(v) = 0
↓(v − v0) = −H−1g(v0)
• Hessian is not positive definite −→ search is not alwaysdownwards
• Much more costly than minima optimization
![Page 89: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/89.jpg)
28
Necessary conditions for a TS optimization
• The starting point must be close to the TS (Pre-optimization: systematic exploration, L(Q)ST,etc)
![Page 90: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/90.jpg)
28
Necessary conditions for a TS optimization
• The starting point must be close to the TS (Pre-optimization: systematic exploration, L(Q)ST,etc)
• Start from an accurately estimated Hessian(analytically or numerically).
![Page 91: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/91.jpg)
28
Necessary conditions for a TS optimization
• The starting point must be close to the TS (Pre-optimization: systematic exploration, L(Q)ST,etc)
• Start from an accurately estimated Hessian(analytically or numerically).
• Some techniques can help in the search (–Eigenvalue Following, Step Walking Surface,NEB–)
![Page 92: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/92.jpg)
29
Choice of the coordinate system
• Good performance ←→ Quadratical behavior.
• Coordinate transformation:
vi = vi(s1, s2, · · · , sM)
• Transformation of F (v) results in a new function F ′:
F ′(s1, s2, . . . , sM) = F (v1(s1, s2, . . . , sM), v2(s1, s2, . . . , sM),. . . , vM(s1, s2, . . . , sM))
• One would like the transformation to give rise to function F ′ thatbehaves quadratically around the critical point.
![Page 93: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/93.jpg)
30
Internal valence coordinates
• Defined in terms of bond distances, bond angles, andbond torsions.
• Expected that the functional dependence on them is moreor less factorized into additive terms.
• The behavior is assumed to be close to quadratic (goodfor optimizations)
• The most widely used → Z-MATRIX
![Page 94: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/94.jpg)
31
Z-Matrix coordinate system
• Ordered definition of the atomic positions in terms ofdistances, angles and torsions with reference to previouslydefined atoms
• Easy way to obtain non-redundant internal valence coor-dinates
• The choice is in general arbitrary
![Page 95: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/95.jpg)
32
Scheme of the Z-matrix logic
����
����
����
����
��
��
��
����
����
����
����
����
����
����
����
����
!!
""##
$$%%
&&''
TREE STRUCTURE
����
����
����
����
��
��
�� ���
�����
���� ?
LOOP STRUCTURE
→ arbitrariness in the definition
![Page 96: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/96.jpg)
33
Internal valence coordinates for periodic systems
• Periodic systems feature in most cases an infinite number ofvalence loops (unless molecular crystals, or 1D polymers)
• The arbitrariness of the Z-Matrix valence coordinates is higherthan in molecules.
![Page 97: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/97.jpg)
34
Coordinate systems suitable for crystals
Cartesian coordinates + lattice parameters
↑ Easy definition
↓ Non quadratic behavior (in general)
↓ When the cell is not kept fixed during optimization → high degreeof dependency between coordinates and lattice parameters
![Page 98: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/98.jpg)
35
Coordinate systems suitable for crystals
Fractionary coordinates + lattice parameters
Atomic coordinates are defined in the basis set of the lattice vectors.
↑ Easy to keep fixed special symmetry positions.
↓ In general, non quadratic behavior
↑ Not too high degree of dependency between coordinates andlattice parameters
![Page 99: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/99.jpg)
36
Dependency of cartesian and fractionary systems on the latticeparameters
FRACT
CART
![Page 100: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/100.jpg)
37
Coordinate systems suitable for crystals
Redundant internal coordinates
• All valence coordinates (bond distances, angles and torsions) areconsidered to form a set of redundant coordinates.
• Hessian, gradient and geometry displacements are built in termsof the redundant coordinates.
• The redundancies are eliminated to obtain the actual geometry(in the cartesian non-redundant space) using numerical approxi-mations
![Page 101: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/101.jpg)
38
Coordinate systems suitable for crystals
Redundant internal coordinates
↑ Quadratic behavior.
↑ Easy choice of the geometrical parameters
↑ Easy to constrain “chemical” degrees of freedom (bond length,angles or dihetrals)
(!!) control optimization for reactivity studies
↓ The size of the redundant space may be very large
↓ The back-transformation from the redundant to the non-redundant(real) space, is a very tricky task.
![Page 102: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/102.jpg)
39
Comparison between redundant and fractionary+latt coordinatesystems. Schlegel Model Hessian
Compound Dim Hamil Num Steps Ered - Efr−l
Nred Nfr−l (µHartree)ZrO2(cubic) 3D PBE0 3 5 -0.2TiO2 (rutile) 3D PW91 5 5 0.0α-quartz 3D B3LYP 7 8 -0.2β-quartz 3D LDA 7 7 0.1Si-Faujasite 3D PBE0 16 17 -1.7Edingtonite (100)+NH3 2D B3LYP 8 10 -18.1Corundum(001)12 layers 2D B3LYP 21 18 -6240.9NaNO2 3D B3PW91 18 18 -0.0CaCO3 3D B3PW91 7 8 0.3ZnGeP2 3D LDA 6 4 -361.5Formaldehyde 3D B3LYP 17 19 1.9Oxalic Acid 3D B3LYP 25 56 4.7Ice 3D PW91 12 19 1.4Li-dopped PA 1D B3LYP 12 14 0.3H2O polymer 1D PBE0 18 19 -0.6
![Page 103: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/103.jpg)
40
Optimization in Crystal2006. Implementation.
![Page 104: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/104.jpg)
40
Optimization in Crystal2006. Implementation.
• Analitic gradient cell+atoms
![Page 105: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/105.jpg)
40
Optimization in Crystal2006. Implementation.
• Analitic gradient cell+atoms
? Several Hamiltonians (HF and DFT)
? Periodicity 0D, 1D, 2D and 3D at the same level of theory.
? All-electron or pseudopotentials.
![Page 106: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/106.jpg)
40
Optimization in Crystal2006. Implementation.
• Analitic gradient cell+atoms
? Several Hamiltonians (HF and DFT)
? Periodicity 0D, 1D, 2D and 3D at the same level of theory.
? All-electron or pseudopotentials.
• Coordinate systems
![Page 107: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/107.jpg)
40
Optimization in Crystal2006. Implementation.
• Analitic gradient cell+atoms
? Several Hamiltonians (HF and DFT)
? Periodicity 0D, 1D, 2D and 3D at the same level of theory.
? All-electron or pseudopotentials.
• Coordinate systems
? Choices
♣ Cartesian/fixed cell
♣ Fractionary+cell
♣ redundant internal valence coordinates
![Page 108: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/108.jpg)
40
Optimization in Crystal2006. Implementation.
• Analitic gradient cell+atoms
? Several Hamiltonians (HF and DFT)
? Periodicity 0D, 1D, 2D and 3D at the same level of theory.
? All-electron or pseudopotentials.
• Coordinate systems
? Choices
♣ Cartesian/fixed cell
♣ Fractionary+cell
♣ redundant internal valence coordinates
? Constraints
♣ some atoms fixed
♣ some cell parameters fixed
♣ fix cell shape (volume optimization or partial fixing) + atoms
♣ fix symmetrized fractionary coordinates
♣ fix internal valence coordinates
♣ symmetry constraints
![Page 109: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/109.jpg)
40
Optimization in Crystal2006. Implementation.
• Analitic gradient cell+atoms
? Several Hamiltonians (HF and DFT)
? Periodicity 0D, 1D, 2D and 3D at the same level of theory.
? All-electron or pseudopotentials.
• Coordinate systems
? Choices
♣ Cartesian/fixed cell
♣ Fractionary+cell
♣ redundant internal valence coordinates
? Constraints
♣ some atoms fixed
♣ some cell parameters fixed
♣ fix cell shape (volume optimization or partial fixing) + atoms
♣ fix symmetrized fractionary coordinates
♣ fix internal valence coordinates
♣ symmetry constraints
• Different Hessian updating schemes: Berny, BFGS, Powell,. . .
![Page 110: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/110.jpg)
40
Optimization in Crystal2006. Implementation.
• Analitic gradient cell+atoms
? Several Hamiltonians (HF and DFT)
? Periodicity 0D, 1D, 2D and 3D at the same level of theory.
? All-electron or pseudopotentials.
• Coordinate systems
? Choices
♣ Cartesian/fixed cell
♣ Fractionary+cell
♣ redundant internal valence coordinates
? Constraints
♣ some atoms fixed
♣ some cell parameters fixed
♣ fix cell shape (volume optimization or partial fixing) + atoms
♣ fix symmetrized fractionary coordinates
♣ fix internal valence coordinates
♣ symmetry constraints
• Different Hessian updating schemes: Berny, BFGS, Powell,. . .
• Starting Model Hessians: Lindh, Schlegel.
![Page 111: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/111.jpg)
41
Structure of Si-Octadecasyl (AST)
![Page 112: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/112.jpg)
42
F− occluded in a D4R unit in as-synthesized zeolites
![Page 113: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/113.jpg)
43
Seeking for a mechanism of F− elimination
• Octadecasyl has been chosen because of a previous experimental work (Villaes-cusa et al, 1998)
• The unit cell consists of 30 atoms
• Cell parameters has been kept fixed in the experimental values
• Atomic positions were fully optimized; all stationary points has been charac-terized as minima or transition states by means of the ab initio vibrationalanalysis
• Methodological level: B3LYP/DZVP//TZVP.
• All energies corrected by ZPE (at DZVP level) and BSSE (at TZVP)
• The starting point of the path has been chosen to be the protonated F-D4Runit, as it is assumed these species are present at the final steps of the templatedecomposition.
![Page 114: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/114.jpg)
44
Protonated F-D4R: reactant
![Page 115: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/115.jpg)
45
Protonated F-D4R: transition state
![Page 116: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/116.jpg)
46
Protonated F-D4R: product
![Page 117: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/117.jpg)
47
The role of a water molecule
![Page 118: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/118.jpg)
48
HF elimination and Si-O-Si bridge condensation: transition state
![Page 119: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/119.jpg)
49
HF elimination and Si-O-Si bridge condensation: product
![Page 120: Strategies in geometry optimization of solids · Strategies in geometry optimization of solids Claudio M. Zicovich-Wilson Facultad de Ciencias, U. A. del Edo. de Morelos, M´exico](https://reader035.vdocuments.us/reader035/viewer/2022071011/5fc9c4edc0037820886a005d/html5/thumbnails/120.jpg)
50
Periodic B3LYP reaction profile for the F− elimination (energiesin kJ/mol)