Quantum Chemistry in the Age of

Machine Learning Pavlo O. Dral

Visit to download the full and correct content document:
https://ebookmass.com/product/quantum-chemistry-in-the-age-of-machine-learning-p
avlo-o-dral/
QUANTUM CHEMISTRY IN THE AGE OF
MACHINE LEARNING
This page intentionally left blank
 QUANTUM
CHEMISTRY IN THE
AGE OF MACHINE
LEARNING
Edited by

PAVLO O. DRAL
Professor, State Key Laboratory of Physical Chemistry of Solid Surfaces, Fujian Provincial Key
Laboratory of Theoretical and Computational Chemistry, Department of Chemistry, College of Chemistry and
Chemical Engineering, Xiamen University, China
Elsevier
Radarweg 29, PO Box 211, 1000 AE Amsterdam, Netherlands
The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United Kingdom
50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States
Copyright © 2023 Elsevier Inc. All rights reserved.
No part of this publication may be reproduced or transmitted in any form or by any means, electronic or
mechanical, including photocopying, recording, or any information storage and retrieval system, without
permission in writing from the publisher. Details on how to seek permission, further information about the
Publisher’s permissions policies and our arrangements with organizations such as the Copyright Clearance
Center and the Copyright Licensing Agency, can be found at our website: www.elsevier.com/permissions.
This book and the individual contributions contained in it are protected under copyright by the Publisher (other
than as may be noted herein).

Notices
Knowledge and best practice in this field are constantly changing. As new research and experience broaden our understanding,
changes in research methods, professional practices, or medical treatment may become necessary.
Practitioners and researchers must always rely on their own experience and knowledge in evaluating and using any information,
methods, compounds, or experiments described herein. In using such information or methods they should be mindful of
their own safety and the safety of others, including parties for whom they have a professional responsibility.
To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors, assume any liability for any
injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or
operation of any methods, products, instructions, or ideas contained in the material herein.

ISBN: 978-0-323-90049-2

For information on all Elsevier publications

visit our website at https://www.elsevier.com/books-and-journals

Publisher: Susan Dennis

Acquisitions Editor: Charles Bath
Editorial Project Manager: Howi De Ramois
Production Project Manager: Paul Prasad Chandramohan
Cover Designer: Matthew Limbert
Typeset by STRAIVE, India
 Contents

Companion website ix Methods 71

Case studies 87
Contributors xi Conclusions and outlook 89
Preface xv Acknowledgments 90
References 90
1
Introduction 4. From small molecules to solid-state
materials: A brief discourse on an example
1. Very brief introduction to quantum of carbon compounds
chemistry Bili Chen, Leyuan Cui, Shuai Wang, and Gang Fu
Xun Wu and Peifeng Su
Introduction 93
Introduction—The foundations of quantum Methods 95
chemistry 3 Case studies 108
Methods of molecular electronic structure Conclusions and outlook 114
computations 8 Acknowledgments 114
Methods of conceptional interpretation based on References 114
electronic structure calculations 20
Case studies 21 5. Basics of dynamics
Conclusions and outlook 24
Xinxin Zhong and Yi Zhao
Acknowledgments 24
References 24 Introduction 117
Methods 119
2. Density-functional theory Case studies 125
Hong Jiang and Huai-Yang Sun Conclusions and outlook 131
Acknowledgments 131
Introduction 27
References 131
Theoretical foundations of DFT 28
Density-functional approximations 34
Practical aspects of DFT implementations 50 6. Machine learning:
Case studies 53 An overview
Concluding remarks 60 Eugen Hruska and Fang Liu
Acknowledgments 61
References 61 Introduction 135
Methods 136
3. Semiempirical quantum mechanical Basic concepts of machine learning 143
methods Case study 147
Pavlo O. Dral and Jan Řezác Conclusions and outlook 149
Acknowledgment 149
Introduction 67 References 149

v
vi Contents

7. Unsupervised learning 2
Rose K. Cersonsky and Sandip De
Machine learning potentials
Introduction 153
Notation guide 154 11. Potentials based on linear models
Methods 155 Gauthier Tallec, Gaetan Laurens, Owen Fresse-Colson, and
Case studies 171 Julien Lam
Conclusions 178
References 178 Introduction 253
Methods 255
Case studies 272
8. Neural networks Conclusion and outlook 276
Pavlo O. Dral, Alexei A. Kananenka, Fuchun Ge, Acknowledgments 276
and Bao-Xin Xue
References 277
Introduction 183
Methods 184 12. Neural network potentials
Case study 200 Jinzhe Zeng, Liqun Cao, and Tong Zhu
Conclusions and outlook 201
Acknowledgments 202 Introduction 279
References 203 Methods 281
Case studies 287
Conclusions and outlook 290
9. Kernel methods Acknowledgment 290
Max Pinheiro Jr and Pavlo O. Dral References 290

Introduction 205 13. Kernel method potentials

Methods 207
Yi-Fan Hou and Pavlo O. Dral
Case studies 227
Conclusions and outlook 229 Introduction 295
Acknowledgments 230 Methods 296
Author contributions 230 Case study 309
References 230 Conclusions and outlook 310
Acknowledgments 311
10. Bayesian inference References 311
Wei Liang and Hongsheng Dai
14. Constructing machine learning
Introduction 233 potentials with active learning
Basic concepts of Bayesian statistics 234 Cheng Shang and Zhi-Pan Liu
Bayesian regression 239
Bayesian inference in machine learning: Bayesian Introduction 313
neural networks 246 Methods 317
Case study 246 Case study 322
Conclusions and outlook 249 Conclusion and outlook 325
Acknowledgments 249 Acknowledgments 325
References 249 References 325
 Contents vii
15. Excited-state dynamics with Conclusions and outlook 449
machine learning Acknowledgments 450
References 450
Lina Zhang, Arif Ullah, Max Pinheiro Jr, Pavlo O. Dral,
and Mario Barbatti
19. Learning dipole moments and
Introduction 329
Methods 332 polarizabilities
Case studies 344 Yaolong Zhang, Jun Jiang, and Bin Jiang
Conclusions and outlook 349
Acknowledgments 350 Introduction 453
References 350 Methods 455
Case studies 461
Conclusions and outlook 463
16. Machine learning for vibrational
Acknowledgments 463
spectroscopy References 464
Sergei Manzhos, Manabu Ihara, and Tucker Carrington

Introduction 355 20. Learning excited-state properties

Methods 357 Julia Westermayr, Pavlo O. Dral, and Philipp Marquetand
Case studies 372
Conclusions and outlook 383 Introduction 467
References 384 Methods 470
Case studies 477
17. Molecular structure optimizations Conclusions and outlook 484
Acknowledgments 486
with Gaussian process regression
References 486
Roland Lindh and Ignacio Fdez. Galván

Introduction 391
Methods 392
4
Case studies 411 Machine learning-improved
Conclusions and outlook 423 quantum chemical methods
Acknowledgments 423
References 424 21. Learning from multiple
quantum chemical methods:
3 Δ-learning, transfer learning, co-kriging,
and beyond
Machine learning of quantum Pavlo O. Dral, Tetiana Zubatiuk, and Bao-Xin Xue
chemical properties
Introduction 491
18. Learning electron densities Methods 493
Bruno Cuevas-Zuvirı́a Case studies 503
Conclusions and outlook 505
Introduction 431 Acknowledgments 506
Methods 435 Author contributions 506
Case studies 445 References 506
viii Contents

22. Data-driven acceleration of Case studies 605

coupled-cluster and perturbation theory Conclusions and outlook 612
Acknowledgments 614
methods
References 614
Grier M. Jones, P.D.Varuna S. Pathirage, and
Konstantinos D. Vogiatzis

Introduction 510
5
Methods 511 Analysis of big data
Case studies 523
Conclusions and outlook 527 26. Analysis of nonadiabatic molecular
Acknowledgment 527
dynamics trajectories
References 528
Yifei Zhu, Jiawei Peng, Hong Liu, and Zhenggang Lan

23. Redesigning density functional Introduction 619

theory with machine learning Theoretical methods 621
Jiang Wu, Guanhua Chen, Jingchun Wang, and Xiao Zheng Examples 630
Case studies 642
Introduction 531 Conclusions and outlook 648
Methods 532 References 649
Case study 552
Conclusions and outlook 554
References 554 27. Design of organic materials with
tailored optical properties: Predicting
24. Improving semiempirical quantum
quantum-chemical polarizabilities and
mechanical methods with machine learning
derived quantities
Pavlo O. Dral and Tetiana Zubatiuk
Gaurav Vishwakarma, Aditya Sonpal, Aatish Pradhan,
Mojtaba Haghighatlari, Mohammad Atif Faiz Afzal, and
Introduction 559
Johannes Hachmann
Methods 560
Case study 572 Introduction 654
Conclusions and outlook 573 Methods 658
Acknowledgments 574 Case studies: Implementing the rational
References 574 design protocol 662
Conclusions and outlook 669
25. Machine learning wavefunction References 670
Stefano Battaglia

Introduction 577 Index 675

Methods 581
 Companion website

Please visit the below companion site to view repositories with programs, data, instructions,
sample input and output files required for case studies as well as any post-publication updates.
https://www.elsevier.com/books-and-journals/book-companion/9780323900492

ix
This page intentionally left blank
 Contributors

Mohammad Atif Faiz Afzal Department of University of Wisconsin-Madison, Madison,

Chemical and Biological Engineering, WI, United States
University at Buffalo, The State University of Leyuan Cui State Key Laboratory for Physical
New York, Buffalo, NY, United States Chemistry of Solid Surfaces, Fujian Provincial
Mario Barbatti Aix Marseille University, CNRS, Key Laboratory of Theoretical and
ICR, Marseille; Institut Universitaire de France, Computational Chemistry, and College of
Paris, France Chemistry and Chemical Engineering,
Stefano Battaglia Department of Chemistry— Xiamen University, Xiamen, China
BMC, Uppsala University, Uppsala, Sweden Hongsheng Dai Department of Mathematical
Liqun Cao Shanghai Engineering Research Sciences, University of Essex, Colchester,
Center of Molecular Therapeutics & New United Kingdom
Drug Development, School of Chemistry and Sandip De BASF SE, Ludwigshafen am Rhein,
Molecular Engineering, East China Normal Germany
University; NYU-ECNU Center for Pavlo O. Dral State Key Laboratory of Physical
Computational Chemistry at NYU Shanghai, Chemistry of Solid Surfaces, Fujian Provincial
Shanghai, China Key Laboratory of Theoretical and
Tucker Carrington Chemistry Department, Computational Chemistry, Department of
Queen’s University, Kingston, ON, Canada Chemistry, and College of Chemistry and
Rose K. Cersonsky Laboratory of Computa- Chemical Engineering, Xiamen University,
tional Science and Modeling, Swiss Federal Xiamen, Fujian, China
Institute of Technology (EPFL), Lausanne, Ignacio Fdez. Galván Department of
Switzerland Chemistry—BMC, Uppsala University,
Bili Chen State Key Laboratory for Physical Uppsala, Sweden
Chemistry of Solid Surfaces, Fujian Provincial Owen Fresse-Colson CEMES, CNRS and
Key Laboratory of Theoretical and Universite de Toulouse, Toulouse Cedex, France
Computational Chemistry, and College of
Gang Fu State Key Laboratory for Physical
Chemistry and Chemical Engineering,
Chemistry of Solid Surfaces, Fujian Provincial
Xiamen University, Xiamen, China
Key Laboratory of Theoretical and
Guanhua Chen Hong Kong Quantum AI Lab Computational Chemistry, and College of
and Department of Chemistry, The University Chemistry and Chemical Engineering,
of Hong Kong, Pokfulam, Hong Kong Xiamen University, Xiamen, China
Bruno Cuevas-Zuvirı́a Centro de Biotecnologı́a Fuchun Ge State Key Laboratory of Physical
y Genómica de Plantas (CBGP, UPM-INIA), Chemistry of Solid Surfaces, Fujian Provincial
Universidad Politecnica de Madrid (UPM), Key Laboratory of Theoretical and
Instituto Nacional de Investigación y Computational Chemistry, Department of
Tecnologı́a Agraria y Alimentaria (INIA), Chemistry, and College of Chemistry and
Madrid, Spain; NASA Center for Early Life Chemical Engineering, Xiamen University,
and Evolution; Department of Bacteriology, Xiamen, Fujian, China

xi
xii Contributors
Johannes Hachmann Department of Chemical
and Biological Engineering, University at
Buffalo; Computational and Data-Enabled
Science and Engineering Graduate Program,
University at Buffalo, The State University of
New York; New York State Center of
Excellence in Materials Informatics, Buffalo,
NY, United States
Mojtaba Haghighatlari Department of
Chemical and Biological Engineering,
University at Buffalo, The State University
of New York, Buffalo, NY, United States
Yi-Fan Hou State Key Laboratory of Physical
Chemistry of Solid Surfaces, Fujian Provincial
Key Laboratory of Theoretical and
Computational Chemistry, Department of
Chemistry, and College of Chemistry and
Chemical Engineering, Xiamen University,
Xiamen, Fujian, China
Eugen Hruska Department of Chemistry,
Emory University, Atlanta, GA, United States
Manabu Ihara Department of Chemical Science
and Engineering, Tokyo Institute of
Technology, Tokyo, Japan
Bin Jiang Hefei National Research Center for
Physical Sciences at the Microscale,
Department of Chemical Physics, University
of Science and Technology of China, Hefei,
Anhui, China
Hong Jiang Beijing National Laboratory for
Molecular Sciences, Institute of Theoretical
and Computational Chemistry, College of
Chemistry and Molecular Engineering, Peking
University, Beijing, China
Jun Jiang Hefei National Research Center for
Physical Sciences at the Microscale,
Department of Chemical Physics, University
of Science and Technology of China, Hefei,
Anhui, China
Grier M. Jones Department of Chemistry,
University of Tennessee, Knoxville, TN,
United States
Alexei A. Kananenka Department of Physics
and Astronomy, University of Delaware,
Newark, DE, United States
Julien Lam CEMES, CNRS and Universite de
Toulouse, Toulouse Cedex, France
Zhenggang Lan South China Normal
University, Guangzhou, China
etan Laurens Institut Lumière Matière,
Ga
UMR5306 Universite Lyon 1-CNRS, Universite
de Lyon, Villeurbanne Cedex, France
Wei Liang School of Mathematical Sciences,
Xiamen University, Xiamen, China
Roland Lindh Department of Chemistry—
BMC, Uppsala University, Uppsala, Sweden
Fang Liu Department of Chemistry, Emory
University, Atlanta, GA, United States
Hong Liu South China Normal University,
Guangzhou, China
Zhi-Pan Liu Collaborative Innovation Center of
Chemistry for Energy Material, Shanghai Key
Laboratory of Molecular Catalysis and
Innovative Materials, Key Laboratory of
Computational Physical Science (Ministry of
Education), Department of Chemistry, Fudan
University; Shanghai Qi Zhi Institute,
Shanghai, China
Sergei Manzhos Department of Chemical
Science and Engineering, Tokyo Institute of
Technology, Tokyo, Japan
Philipp Marquetand Institute of Theoretical
Chemistry, Faculty of Chemistry, University
of Vienna, Vienna, Austria
Jiawei Peng South China Normal University,
Guangzhou, China
Max Pinheiro Jr Aix Marseille University,
CNRS, ICR, Marseille, France
Aatish Pradhan Department of Chemical and
Biological Engineering, University at Buffalo,
The State University of New York, Buffalo,
NY, United States
Jan Řezác Institute of Organic Chemistry and
Biochemistry of the Czech Academy of
Sciences, Prague, Czech Republic
P.D.Varuna S. Pathirage Department of
Chemistry, University of Tennessee,
Knoxville, TN, United States
Cheng Shang Collaborative Innovation Center
of Chemistry for Energy Material, Shanghai
Key Laboratory of Molecular Catalysis and
Innovative Materials, Key Laboratory of
Computational Physical Science (Ministry of
 Contributors xiii
Education), Department of Chemistry, Fudan University of Warwick, Coventry, United
University; Shanghai Qi Zhi Institute, Kingdom
Shanghai, China Jiang Wu Hong Kong Quantum AI Lab and
Aditya Sonpal Department of Chemical and Department of Chemistry, The University of
Biological Engineering, University at Buffalo, Hong Kong, Pokfulam, Hong Kong
The State University of New York, Buffalo, Xun Wu The State Key Laboratory of Physical
NY, United States Chemistry of Solid Surfaces, Fujian Provincial
Peifeng Su The State Key Laboratory of Key Laboratory of Theoretical and
Physical Chemistry of Solid Surfaces, Fujian Computational Chemistry, and College of
Provincial Key Laboratory of Theoretical and Chemistry and Chemical Engineering,
Computational Chemistry, and College of Xiamen University, Xiamen, Fujian, China
Chemistry and Chemical Engineering, Bao-Xin Xue State Key Laboratory of Physical
Xiamen University, Xiamen, Fujian, China Chemistry of Solid Surfaces, Fujian Provincial
Huai-Yang Sun Beijing National Laboratory for Key Laboratory of Theoretical and
Molecular Sciences, Institute of Theoretical and Computational Chemistry, Department of
Computational Chemistry, College of Chemistry, and College of Chemistry and
Chemistry and Molecular Engineering, Peking Chemical Engineering, Xiamen University,
University, Beijing, China Xiamen, Fujian, China
Gauthier Tallec Institut des systèmes Jinzhe Zeng Shanghai Engineering
intelligents et de robotique (ISIR), Sorbonne Research Center of Molecular Therapeutics &
Universite, Paris, France New Drug Development, School of
Arif Ullah State Key Laboratory of Physical Chemistry and Molecular Engineering,
Chemistry of Solid Surfaces, Fujian Provincial East China Normal University; NYU-ECNU
Key Laboratory of Theoretical and Center for Computational Chemistry at NYU
Computational Chemistry, Department of Shanghai, Shanghai, China
Chemistry, and College of Chemistry and Lina Zhang State Key Laboratory of Physical
Chemical Engineering, Xiamen University, Chemistry of Solid Surfaces, Fujian Provincial
Xiamen, Fujian, China Key Laboratory of Theoretical and
Gaurav Vishwakarma Department of Chemical Computational Chemistry, Department of
and Biological Engineering, University at Chemistry, and College of Chemistry and
Buffalo, The State University of New York, Chemical Engineering, Xiamen University,
Buffalo, NY, United States Xiamen, Fujian, China
Konstantinos D. Vogiatzis Department of Yaolong Zhang Hefei National Research Center
Chemistry, University of Tennessee, for Physical Sciences at the Microscale,
Knoxville, TN, United States Department of Chemical Physics, University
of Science and Technology of China, Hefei,
Jingchun Wang Department of Chemical
Anhui, China
Physics, University of Science and Technology
of China, Hefei, China Yi Zhao State Key Laboratory of Physical
Chemistry of Solid Surfaces, Fujian Provincial
Shuai Wang State Key Laboratory for Physical
Key Laboratory of Theoretical and
Chemistry of Solid Surfaces, Fujian Provincial
Computational Chemistry, College of
Key Laboratory of Theoretical and
Chemistry and Chemical Engineering,
Computational Chemistry, and College of
Xiamen University, Xiamen, China
Chemistry and Chemical Engineering,
Xiamen University, Xiamen, China Xiao Zheng Department of Chemical Physics,
University of Science and Technology of
Julia Westermayr Theoretical Surface
China, Hefei, China
Chemistry Group, Department of Chemistry,
xiv Contributors

Xinxin Zhong Collaborative Innovation Center Molecular Engineering, East China Normal
for Advanced Organic Chemical Materials University; NYU-ECNU Center for
Co-constructed by the Province and Ministry, Computational Chemistry at NYU Shanghai,
Ministry of Education Key Laboratory for the Shanghai, China
Synthesis and Application of Organic Yifei Zhu South China Normal University,
Functional Molecules, Hubei University, Guangzhou, China
Wuhan, China
Tetiana Zubatiuk Department of Chemistry,
Tong Zhu Shanghai Engineering Research Carnegie Mellon University, Pittsburgh, PA,
Center of Molecular Therapeutics & New United States
Drug Development, School of Chemistry and
 Preface
Quantum chemistry is a very mature field
of science—its method development started
as early as 1926, when the Schr€ odinger
equation was proposed and, a year later, ap-
plied to solving the first molecular system
H2. Nevertheless, the quick and astonishing
progress in another discipline—artificial in-
telligence and machine learning—has had a
big impact on how we perform quantum
chemical simulations. Machine learning in
quantum chemistry has emerged as a vi-
brant and ever-changing field rich in many
methodological advantages and has rapidly
become a commonly used technique in com-
putational chemistry. This calls for a book
that can be used as a textbook in advanced
courses on theoretical and computational
chemistry as well as a reference material
for specialists and nonspecialists alike.
Teaching machine learning methods in
quantum chemistry has become a must in
the university-level courses on theoretical
and computational chemistry as I have ar-
gued in my Perspective article with the same
title as this book in early 2020 [1]. I have
started to teach the very basics of machine
learning topics myself in the spring semester
of 2021 for undergraduate chemistry stu-
dents at Xiamen University. My lecture notes
turned out to be very useful when contribut-
ing to some of the chapters. The aforemen-
tioned Perspective has also pretty much
formed the blueprint for this book, largely
defining the scope of the topics covered,
and the reader may refer to this Perspective
for a very concise, bird’s-eye view of ma-
chine learning in quantum chemistry.
xv
Part 1 of this book is dedicated to introduc-
ing the minimum of fundamental topics in
both quantum chemistry and machine learn-
ing, which are indispensable to lay the foun-
dation for the other parts; thus, Part 1 with 10
chapters is the largest in the book. Quantum
chemistry topics start with a brief introduc-
tion to quantum chemistry in Chapter 1,
followed by density functional theory in
Chapter 2, semiempirical quantum mechani-
cal methods in Chapter 3, and a brief dis-
course dealing with quantum chemical
calculations of systems from molecules to
solid-state materials in Chapter 4. Chapter 5
covers an important introduction to dynam-
ics simulations. Machine learning is intro-
duced in Chapter 6 followed by chapters
covering specific key topics such as
unsupervised learning in Chapter 7, neural
networks in Chapter 8, kernel methods in
Chapter 9, and Bayesian inference in
Chapter 10.
Part 2 is dedicated to one of the most
abundant applications of machine learning
when it is used as a surrogate for quantum
chemical potentials; by potential we mean
the atomistic system total energy as a func-
tion of nuclear coordinates. Chapter 11
covers potentials based on linear models
followed by Chapters 12 and 13 introducing
neural network potentials and kernel
method potentials, respectively. Chapter 14
covers the important topic of creating train-
ing data for such potentials via active learn-
ing. Chapter 15 focuses on the challenges in
creating machine learning models for
performing excited-state dynamics. Machine
xvi
learning also helps in vibrational spectros-
copy as discussed in Chapter 16 and in accel-
erating molecular structure optimizations
with quantum chemical methods as shown
in Chapter 17.
Part 3 provides a taste of how machine
learning can be used to learn other important
quantum chemical properties such as elec-
tron densities (Chapter 18), dipole moments,
and polarizabilities (Chapter 19), and
excited-state properties (Chapter 20).
Part 4 describes how machine learning be-
comes an integral part of quantum chemical
simulations by improving quantum chemical
methods, e.g., via general approaches such
as Δ-learning (Chapter 21, which also covers
related topics of learning from multiple quan-
tum chemical methods such as transfer learn-
ing, co-kriging, hierarchical machine learning,
and their combinations) or dedicated ap-
proaches for accelerating coupled-cluster
and perturbation theory quantum chemical
methods (Chapter 22), redesigning density
functional theory methods (Chapter 23), and
improving semiempirical quantum chemical
methods (Chapter 24). Part 4 concludes
by overviewing how machine learning can
learn the essence of quantum chemistry—
wavefunction—in Chapter 25.
Part 5 concludes the book with examples
of analysis of big data in a quantum chemical
context, showing specific examples for ana-
lyzing nonadiabatic molecular dynamics
trajectories (Chapter 26) and searching for
organic materials with tailored optical prop-
erties in the wide sea of possible compounds
(Chapter 27).
One of the main motivations of this book
has been to offer purely theoretical knowl-
edge alongside practical knowledge by pro-
viding mathematical expositions in the
Methods section and incorporating computa-
tional and, sometimes, pen-and-paper exer-
cises at the end of each chapter in the Case
Preface
studies section. The case studies in many
chapters revolve around one of the simplest
chemical systems, the H2 molecule, which is
intentionally chosen to demonstrate the basic
ideas of various methods to ensure that they
are well understood before moving on to
more complex examples. Many chapters
also contain examples from modern studies
so that the reader can better understand
more advanced concepts. The reader is al-
ways encouraged to go beyond the instruc-
tions in the case studies and experiment on
their own.
This broadness of the topics could not
be covered by a single person within a
reasonably short amount of time adequate
for such a fast-paced field, especially as
many colleges have already introduced
machine learning-related topics for chem-
ists and we have needed a textbook pref-
erably yesterday. Thus, this book is a
product of a massive effort by 65 authors
from very different backgrounds ranging
from mathematics to physics, materials
science, and chemistry. Many of the au-
thors are among the best specialists in
the topics they covered, which is evident
from the fact that many of them were
speakers at the first International Sympo-
sium on Machine Learning in Quantum
Chemistry that was held online from No-
vember 12 to 14, 2021.
This is only the tip of the iceberg, as many
more people were peer reviewing and all of
my group members (Lina Zhang, Arif Ullah,
Bao-Xin Xue, Yi-Fan Hou, Fuchun Ge,
Peikun Zheng, Shuang Zhang, Wudi Yang,
Ran Wang, Linqiang Wei) were heavily in-
volved in creating this book by becoming
co-authors, additional peer reviewers, and
testers for the case studies. The publisher’s
team (Anneka Hess, Devlin Person, Jai Marie
Jose, Howell Angelo M. De Ramos, Paul
Prasad Chandramohan, Indhumathi Mani,
 Preface xvii
and many more behind the scenes), were in- and publisher’s team for their efforts that
strumental to accomplish this project from made this book possible.
initiation to finalizing and supporting it all Happy reading and learning!
the way through. Just to give a short perspec-
tive on the amount of effort put into this Pavlo O. Dral
book: from the suggestion to write the book Xiamen, China
on May 13, 2020 (by Anneka Hess from
Elsevier) until writing this Preface, my email
folder for this project contains around 2000
emails; this is not including all the discus-
sions via messaging apps and meetings to co- Reference
ordinate our efforts. I cordially thank all the
[1] P.O. Dral, Quantum chemistry in the age of machine
authors, peer reviewers, my group members,
learning, J. Phys. Chem. Lett. 11 (2020) 2336–2347.
This page intentionally left blank
 P A R T 1

Introduction
This page intentionally left blank
 C H A P T E R

1
Very brief introduction to quantum
chemistry
Xun Wu and Peifeng Su
The State Key Laboratory of Physical Chemistry of Solid Surfaces, Fujian Provincial Key
Laboratory of Theoretical and Computational Chemistry, and College of Chemistry and Chemical
Engineering, Xiamen University, Xiamen, Fujian, China

Abstract
This chapter provides a brief introduction to quantum chemistry including the basic concepts and approxima-
tions of quantum mechanics, and a series of quantum chemical methods.

Keywords: Quantum mechanics, Quantum chemistry, Hartree–Fock method, Configuration interaction,

Møller–Plesset perturbation theory, Coupled-cluster method, Multi-configurational self-consistent-field,
Time-dependent Hartree–Fock

Introduction—The foundations of quantum chemistry

Quantum chemistry (QC) refers to the application of quantum mechanics to study the
chemical and physical properties of atoms, molecules, and materials. It is generally accepted
that the first QC calculation was performed by the German physicists Walter Heitler and Fritz
London on the hydrogen (H2) molecule in 1927. Now, QC has become an irreplaceable tool for
chemists to explore fundamental scientific questions in chemistry and related fields, [1–4]
with numerous applications to atomistic systems such as atoms, molecules or ionic solids.
This chapter introduces basic knowledge of quantum mechanics and QC methods for solving
the Schr€odinger equation.

Brief introduction of quantum mechanics

Basic concepts
The wave function of a system is a mathematical expression that contains information
about anything that may be measured for this system. In quantum mechanics, the wave

Quantum Chemistry in the Age of Machine Learning 3 Copyright # 2023 Elsevier Inc. All rights reserved.
https://doi.org/10.1016/B978-0-323-90049-2.00006-8
4 1. Very brief introduction to quantum chemistry

function Ψ(r1, r2…rn) is a function of the positions of all the particles in the system, describing
a time-independent quantum mechanical system, in which r1, r2…rn are the positions of N
particles. The square of the wave function, jΨ(r1, r2…rn)j2, represents the probability of find-
ing N particles in a given place.
Every observable physical quantity, such as position, energy, or momentum, is described
Ð Ð  ∗
by an operator. A Hermitian operator A ^ satisfies f ∗ Agdτ
^ ¼ g Af ^ dτ, where f and g are
well-behaved functions. Measurement shows a prediction as a result of an operator acting
^ Hamiltonian.
on a wave function. The most important operator is the total energy operator Hor
The measurement of the Hamiltonian operator is described by an eigen-equation called the
time-independent Schr€ odinger equation. Solutions for the time-independent Schr€ odinger
equation only exist for certain values of energy (energy is quantized in bound states). The
resulting eigenvalue equation is:
^ ðr1 , r2 ⋯rn Þ ¼ EΨðr1 , r2 ⋯rn Þ
HΨ (1)
In Eq. (1), E is the eigenvalue, which is the energy of the system described by Ψ(r1, r2…rn).
Moreover, the wave function must generally depend on time to reproduce information about
the system’s evolution over time. The time-dependent wave function, expressed as Ψ(r1, r2…rn, t),
satisfying the time-dependent Schr€
odinger equation:
∂Ψðr1 , r2 ⋯rn , tÞ ^
iħ ¼ HΨðr1 , r2 ⋯rn , tÞ (2)
∂t

Born–Oppenheimer approximation
The solution of the Schr€
odinger equation for molecules systems is the primary task of QC.
However, an exact solution cannot be obtained for multi-electron systems. For a multi-
electron system (atom or molecule), the time-independent Hamiltonian in atomic units is:
X
N elec
1 2
NX
nuclei
1 2
NX X
nuclei N elec
1 X
N X
elec N elec
1
NX X
nuclei N nuclei
Za Zb
^ ¼
H ri  ra  +   +
2 2 jRa  ri j  rj  ri  jRa  Rb j
i a a i i j>i α b>a

(3)
The Hamiltonian in Eq. (3) contains the electron kinetic operators (the first term), nuclear
kinetic operators (the second term), nucleus-electron attractive potential operators (the third
term), electron–electron repulsion (the fourth term), and nucleus-nucleus repulsion (the final
term). The Born–Oppenheimer (BO) approximation is introduced to simplify the Hamilto-
nian. In the BO approximation, the coupling between the nuclear r2 and the electronic wave
function is neglected. In most cases, the coupling between two electronic states is not strong,
i.e., the energy difference between two electronic states is big enough so that a small displace-
ment of nuclei will not change the electronic state, making the BO approximation valid, which
allows the separation of electrons (light particles) from nuclei (heavy particles) on time scales,
resulting in a considerable simplification.
According to this approximation, the total Hamiltonian operator can be expressed as the
sum of electronic Hamiltonian and nuclear Hamiltonian. The electronic Hamiltonian is writ-
ten as:
 Introduction—The foundations of quantum chemistry 5
X
N elec NX X
nuclei N elec X
N X
elec N elec
^ el ¼  1
H r2i 
1
+ 
1
 (4)
2 jRa  ri j rj  ri 
i¼1 a i i j>i

Unless otherwise specified, the Hamiltonian used in this chapter refers to the electronic
Hamiltonian of Eq. (4).
According to the Pauli principle, the wave function of a multi-electron atom or molecule
should be antisymmetric with respect to the exchange of two electrons. Antisymmetrization
leads to Fermi holes, preventing the close approach of the electrons with the same spin and
reducing the repulsion energy between them. Thus, the many-electron function can be ap-
proximately written as the antisymmetrized product of N one-electron functions in the form
of a Slater determinant (SD):
 
 ϕ1 ðr1 Þ ϕ2 ðr1 Þ ⋯ ϕN ðr1 Þ 
 
1  ϕ1 ðr2 Þ ϕ2 ðr2 Þ ⋯ ϕN ðr2 Þ 
ψ ðr1 , r2 ⋯, rN Þ ¼ pffiffiffiffiffiffi   (5)
N!  ⋮ ⋮ ⋯ ⋮ 
 ϕ ðr Þ ϕ ðr Þ ⋯ ϕ ðr Þ 
1 N 2 N N N

In Eq. (5), ϕ1…ϕN refer to spin orbitals.

The solution of the Schr€odinger equation involves calculations of all pairwise interactions
between electrons (see the last term of electronic Hamiltonian, Eq. (4)) and electrons and nu-
clei, which becomes analytically intractable for more than a two-body problem (one nucleus
and one electron). This leads to the fact that the Schr€
odinger equation cannot be solved exactly
for multi-electron systems. Two approximating approaches, variational principle, and pertur-
bation theory are widely used to solve the Schr€ odinger equation to extract the necessary in-
formation about atoms or molecules of interest.

Variational principle
Given a system whose Hamiltonian is time-independent and whose lowest-energy eigen-
values is E1, if ϕ is any normalized, well-behaved function of coordinates of the system’s par-
ticles that satisfies the boundary conditions, then
ð
^
ϕ∗ Hϕdτ  E1 (6)

If the trial function has a minimum of energy with respect to variational coefficient c, we
can find it by
∂ ∂ D  ^  E
h Ei ¼ ϕHϕ ¼0 (7)
∂c ∂c
The lower energy, the better the trial wave function. And if the trial function is accurate
enough, the function we get will be close enough to the true ground-state wave function.
To improve the quality of trial wave function, the linear variational principle is applied,
by which the trial wave function is constructed by a linear combination of n linearly indepen-
dent functions (χ 1, χ 2, ⋯ χ n):
ϕ ¼ c1 χ 1 + c2 χ 2 +⋯+cn χ n (8)
6 1. Very brief introduction to quantum chemistry

where c1…cn are the variational coefficients to be optimized.

Based on this principle, the minimization routine regulates the parameters to obtain the
wave function corresponding to the minimum energy, which is the wave function that closely
approximates the ground state.
From the variational principle, the value of the lowest root is the upper bound for the sys-
tem’s ground-state energy. With the linear variational principle, instead of obtaining the trial
wave function, the combination coefficients of the basis functions should be solved.

Perturbation theory
Perturbation theory provides a procedure for finding approximate solutions to the
Schr€odinger equation for a system that differs only slightly from a system for which the
solutions are known. This method is frequently employed to estimate the functions or values
based on partial knowledge about the solutions of the investigated problem.
odinger equation for the Hamiltonian operator H
If the Schr€ ^
^ n ¼ En ψ n
Hψ (9)
ð0Þ
^ :
cannot be solved exactly, but we know the exact solutions of the Hamiltonian operator H
^ ð0Þ ψ ð0Þ ¼ Eð0Þ ψ ð0Þ
H (10)
n n n

where the superscript (0) denotes zero-order, which means that the wave function and energy
are already known, the subscript n denotes the nth electronic state.
In perturbation theory, we assume that H^ may be expanded as:

^ ð0Þ + H
^ ¼H
H ^0 (11)
0
The operator H^ is called the perturbation. Since the operator H
^ differs only slightly from
ð0Þ
^ , the eigenfunctions and eigenvalues of H
H ^ will not differ significantly from those of the
ð0Þ
unperturbed Hamiltonian operator H ^ .
If we suggest that the wave function and energy can be expressed as the linear expansion
according to λ, which can be written as
ψ n ¼ ψ ðn0Þ + λψ ðn1Þ + λ2 ψ ðn2Þ +⋯ + λk ψ ðnkÞ +⋯ (12)

En ¼ Eðn0Þ + λEðn1Þ + λ2 Eðn2Þ +⋯ + λk EðnkÞ +⋯ (13)

Here the superscript (k) denotes the order of wave function and energy according to the ex-
pansion. In the limit λ ! 0, the perturbed system reduces to the unperturbed system.
First, it is assumed that the zero-order wave function is not degenerate for simplicity. By
equating the coefficients of the same power of λ, we can obtain a series of equations, from
which we can obtain the first-, second-, and higher-order corrections. The first-order energy
correction is an average of the perturbation operator over the unperturbed wave function
D   E ð
ð0Þ  ^ 0  ð0Þ ^0 ψ ð0Þ dτ
En ¼ ψ n H ψ n ¼ ψ ðn0Þ∗ H
ð1Þ
n (14)
 Introduction—The foundations of quantum chemistry 7
while the second-order energy corrections can be expressed as:
D   E
 ð0Þ  ^ 0  ð0Þ 2
X  ψ m H ψ n 
Eðn2Þ ¼ (15)
m6¼n Eðn0Þ  Eðm0Þ

The first-order wave function correction ψ (1) n is orthogonal to the unperturbed wave
function.
D   E
ð0Þ  ^ 0  ð0Þ
X ψ m H ψ n
ψ ðn1Þ ¼ ð0Þ ð0Þ
ψ ðm0Þ (16)
m6¼n E n  E m

It can be found that ψ (1) n can be expressed as the linear combination of m (m 6¼ n) electronic
th
th th
states. The smaller
D the  energy
 E difference between m state and n state is, or the larger the
ð0Þ  ^ 0  ð0Þ
coupling term ψ m H ψ n is, the larger the combination coefficient of the mth electronic
state wave function could be.
Second, if the zero-order wave function is degenerate, any arbitrary combination of the de-
generate wave functions is the eigenfunction of H^ ð0Þ. However, introducing the perturbation
0
operator H ^ will completely or partially remove the degeneracy. Thus, the most important
issue is to derive the combination coefficients of the degenerate wave function. It is called
a degenerate perturbative theory, which can be seen from the textbook.

Comparison of the variation principle and perturbation theory

The perturbation method applies to all the bound states of a system. The variation theorem
applies to the lowest state of a given symmetry. Also, we can use the linear variation method
to treat the excited bound states. In the perturbation method, one can calculate the energy
much more accurately (to an order of 2 k + 1) than the wave function (to an order of k). The
accuracy of perturbation calculations depends on zero-order Hamiltonian and wave function,
while the variation method can get a relatively good result with a rather inaccurate wave
function.

Fundamentals of quantum chemistry

Based on the fundamental postulates, concepts and approximations in quantum mechan-
ics, QC is devoted to exploring the electronic structure of multi-electron systems by using mo-
lecular electronic structure methods that solve the molecular Schr€ odinger equation associated
with different kinds of molecular Hamiltonian. Methods that do not include any empirical or
semiempirical parameters are called ab initio methods. This does not indicate that their so-
lutions are exact; they are all approximate. It means that a particular approximation is rigor-
ously defined on first principles (quantum theory) and then solved within an error margin
that is qualitatively known beforehand.
8 1. Very brief introduction to quantum chemistry

Categories of molecular electronic structure methods

Molecular electronic structure methods can be commonly divided into two categories, den-
sity functional theory (DFT) and wave function theory (WFT), within WFT, there are also two
categories, valence bond (VB) method and molecular orbital (MO) methods. VB method was
developed by W. Heitler, F. London, J. C. Slater and L. Pauling. One of the significant advan-
tages of VB theory is its visually intuitive wave function. The physical basis of VB theory is
that a chemical bond arises from the spin-exchange between two electrons. Thus, it aims to
construct wave functions in which all possible bonding patterns are described in terms of spin
coupling. As such, VB theory deals with multi-reference wave functions constructed directly
from atomic orbitals (AOs), which are usually non-orthogonal for different atoms, whereas
MO-based methods usually use orthogonal delocalized orbitals. In the past decade, various
accurate VB methods and orbital optimization algorithms have been proposed to improve
computational efficiency and accuracy. Several review articles and books on various VB
methods published in recent years are recommended to interested readers [5,6].
The MO method was introduced in 1929 by Friedrich Hund and Robert S. Mulliken, which
describes electrons by mathematical functions delocalized over an entire molecule. Since the
1960s, the MO method has become an irreplaceable tool for qualitative interpretation of mo-
lecular properties and chemical reactions. In the next section, a series of MO methods will be
introduced.

Methods of molecular electronic structure computations

Hartree–Fock method
The Hartree–Fock (HF) method seeks to approximately solve the electronic Schr€ odinger
equation with the assumption that the wave function can be approximated by a single SD
made up of one set of spin orbitals. The HF method is a wave function-based approach based
on the independent-particle model, which reduces the many-electron problem to indepen-
dent single-electron problems. It means that the interactions between the particles are approx-
imated by taking all interactions into account in an average fashion.
The HF method is derived from the variational principle, which is applied to an energy
functional that includes Lagrange multipliers for orthonormal constraints and defines the
HF matrix for each orbital. First, by using the Condon–Slater rules, the energy of a single de-
terminant can be expressed as:
X n  
n
1X n X
E¼ hii + J ij  Kij (17)
i¼1
2 i¼1 j¼1

where Jij and Kij are Coulomb integral and exchange integral between spin orbitals i and j,
respectively:
ðð
1
J ij ¼ hijjiji ¼ dτ1 dτ2 ϕ∗i ð1Þϕ∗j ð2Þ ϕi ð1Þϕj ð2Þ (18)
r12
 Methods of molecular electronic structure computations 9
ðð
1
Kij ¼ hijjjii ¼ dτ1 dτ2 ϕ∗i ð1Þϕ∗j ð2Þ ϕ ð1Þϕi ð2Þ (19)
r12 j
hii is the one-electron integral
D E ð
 
hii ¼ i^hð1Þi ¼ dτ1 ϕ∗i ð1Þ^hð1Þϕi ð1Þ (20)

where ^hð1Þ is the one-electron operator including electron kinetic operator and electron-
nucleus potential operator shown in Eq. (3).
The objective now is to determine a set of MOs that makes the energy a minimum or at least
stationary with respect to a change in the orbitals. The variation must be carried out to ensure
that MOs remain orthogonal and normalized. Since E varies subjected to a constraint that the
spin orbitals are orthonormal,
hijji  δij ¼ 0 (21)
where hi j ji is the Dirac notation of the inner product of two spin orbitals,
ð
hijji ¼ dτϕ∗i ϕj (22)

and δij is the Kronecker delta,


1, i ¼ j
δij ¼ (23)
0, i 6¼ j
A constrained optimization by using Lagrange multipliers can be expressed as:
0 1
X  
δ @E  εij hijji  δij A ¼ 0 (24)
ij

X 1 X X  XX
δhii + δJ ij  δKij  εij δhijji ¼ 0 (25)
i
2 i j i j

where εij refer to the Lagrange multiplier matrix elements, and the notation δ indicates the
variation of the functional.
The variation of the energy leads to the equation below:
2 3
Xð X  X
δϕ∗i ð1Þ4^
hð1Þϕi ð1Þ + ^ j ð 1Þ ϕ i ð 1Þ 
^J j ð1Þ  K εij ϕj ð1Þ5dτ1 + c:c: ¼ 0 (26)
i j j

In Eq. (26), c.c. is the complex conjugate of the summation ahead. δϕi indicates the variation
of the spin orbital ϕi (ϕi ! ϕi + δϕi). ^J ð1Þ and K
^ ð1Þ operators are connected with the potential
energy of electron 1 with all other electrons in the system. They are defined via the action on
any function of the position of electron 1:
10 1. Very brief introduction to quantum chemistry
X
^J ð1Þ ¼ ^J j ð1Þ
j
ð (27)
^J j ð1Þuð1Þ ¼ dτ2 1 ϕ∗ ð2Þϕj ð2Þuð1Þ
r12 j
X
^ ð1Þ ¼
K ^ j ð 1Þ
K
j
ð (28)
^ j ð1Þuð1Þ ¼ dτ2 1 ϕ∗ ð2Þuð2Þϕj ð1Þ
K
r12 j
According to Eq. (26), we have
  X
^
hð1Þ + ^J ð1Þ  K
^ ð 1Þ ϕ i ð 1Þ ¼ εij ϕj ð1Þ (29)
j

^
hð1Þ + ^J ð1Þ  K
^ ð1Þ is defined as the Fock operator, which is an effective one-electron energy
operator. It describes the kinetic energy of an electron and the attraction to all the nuclei
and the repulsion to all the other electrons.
Eq. (29) may be simplified by choosing a unitary transformation that makes the matrix of
Lagrange multipliers diagonal, that is εij ¼ 0. This particular set of MOs is called canonical
MOs and thus transforms Eq. (29) into a set of pseudo-eigenvalue equations:
^ð1Þϕi ð1Þ ¼ εi ϕi ð1Þ
F (30)
where εi is the orbital energy of the Hartree–Fock orbital.
D E X
 
εi ¼ i^ hð1Þi + ½hijjiji  hijjjii (31)
j

The HF equations form a set of pseudo-eigenvalue equations as the Fock operator depends
on all the occupied MOs (via the Coulomb and exchange operators). A specific Fock operator
can only be determined if all the other occupied orbitals are known, and therefore the self-
consistent-field (SCF) procedure must be employed to solve the problem.
The HF wave function is not unique. The total wave function is unchanged by a unitary
transformation of the occupied MOs among themselves (rows and columns in a determinant
can be added and subtracted without affecting the determinant itself ). The canonical MOs
obtained from the Fock equations can be transformed into localized orbitals with various lo-
calization schemes.
The total HF energy of a molecule is expressed as:
X
n
1Xn X
n
EHF ¼ hii + ½hijjiji  hijjjii + V NN (32)
i
2 i j

where VNN is nuclear-nuclear repulsion energy and is not related to wave function. For an
atom, VNN is zero. Considering Eqs. (31) and (32), it is noticed that the total energy is not sim-
ply a sum of MO orbital energies.
An electron at r is viewed as interacting with an “averaged” charge density, “mean-field”,
arising from the N  1 remaining electrons. Originally, the HF calculations were performed
using numerical methods to solve the HF differential equations, which are complex for
Another random document with
no related content on Scribd:
 Assistiu-lhe o piedoso bispo até o ultimo valle, e logo seu corpo foi
levado por homens principaes ao Hospicio de Nossa Senhora da
Penha dos Capuchinhos Francezes, o dia em que chegavam as
novas da restauração do famoso Palmar a Pernambuco, que havia
de ser o sexto da victoria, pois tanto gasta um caminheiro apressado
de um logar a outro. Mas é em vão busca-lo em Pitta, auctor
moderno que d’isto tracta como si não tractára. E mais me
escandalisa que passasse em sua mesma patria por um poeta de tal
nome seu contemporaneo, com quem devia gastar parte d’aquelles
elogios. Morreu finalmente no anno de 1696 com edade de setenta
e tres annos.
Este é o mais abreviado resumo que posso dar da vida do meu
suspirado, quão dilectissimo poeta lyrico; e oxalá podéra eu publicar
os prodigiosos fundamentos do meu amor, derramando entre as
gentes o manancial thesouro de suas graças! Singular foi a estrella
que dominou em seu engenho; porque a toda a circumferencia das
luzes apolineas brilhou com egualdade senhoril; e não menos
prodigioso aquelle não sei que de sua guarda, porque offendendo ás
claras muitas pessoas, de quem o menor movimento seria sem
duvida uma tyranna morte, sempre se atreveu, e nunca de seu motu
proprio cautelou perigos; morrendo intacto de tão prolongados
mezes.
Muitos eram os feridos do seu ferro que consultaram o remedio no
mesmo instrumento da chaga, beijando a Achilles a lança que os
traspassára. Raro testimunho d’esta fatalidade foi a resposta que
deu a um queixoso certo governador severamente resoluto: «Não
faça V. Mᶜᵉ. caso (disse), porque isso tambem passa por mim, sem
que por mim passe a minima tenção de o castigar.»
Testimunho d’esta fatalidade são as duas quartas de um soneto,
que se fez em sua morte; o qual não escrevo por inteiro em razão
de que si os seus principios professam a verdade pura, os fins
todavia contém temeraria petulancia:

Morreste emfim, Gregorio esclarecido,

Que sabendo tirar por varios modos
 A fama, a honra, o credito de todos,
D’esses mesmos te viste applaudido.

Entendo que outro tal não tem nascido

Entre os Romanos, Gregos, Persas, Godos,
Que comtigo mereça ter apodos
Nos applausos, que assim has adquirido.

Muitas vezes quiz elle refrear o genio, que conhecia

prejudialmente peccaminoso, fazendo os actos de christão que em
seu logar veremos, mas debalde o intentava, porque o seu furor
intrepido imperava dominante na massa sanguinaria contra os
desacertos d’aquella edade, castigados por Deus com tão horrorosa
peste e tão repetidas fomes: como tambem veremos pelo decurso
d’estas obras. E não é de admirar que disparadas do throno da
divina justiça aquellas duas lanças de sua via, seguisse a terceira
com tão exquisito genero de guerra em um homem, que de sua mãe
unicamente tomou este appellido entre outros partos: ella o deu
appelidando-se—da Guerra—, e elle o foi sem aquella proposição
da, por ser a mesma guerra, e não o instrumento d’ella. Isto parece
que prophetizou corto inimigo seu, respondendo-lhe a uma satyra
com outra na seguinte fórma:

Porém si em nada és guerreiro,

Para que te chamas guerra,
E a fazes a toda a terra
Com a lingua, que é mor damno, &.

Deixou o dr. Gregorio de Mattos um filho de sua mulher Maria de

Povos, chamado Gonçalo de Mattos, cujo amor publica em várias
obras este livro, que em seus logares se verão sem enfadosas citas.

o quente da cama
Com Gonçalo, e com sua ama,
 Dizendo estava comei-me, &.
Por vida do meu Gonçalo,
Custodia formosa e linda, &.
Madrasta do Gonçalinho,
Que é lindo enteado a fé, &.
Sim, por vida de Gonçalo, &.
Mas por vida de Gonçalo, &.

D’este moço, que com sua mãe ficou em summa pobreza e

desamparo, correm noticias muito geraes que totalmente
degenerára d’aquella massa scientifica de seus estupendos
progenitores. Bem pudera eu duvida-lo em uma terra, onde sempre
se hão de tomar os echos da fama pelo contrario; pois nunca vi nella
abonar um sujeito que não mereça ser desterrado por máu, nem
vituperar outro que ao contrario desmereça elogios de bom.
Mas para cumprir com os relativos d’esta historia consultei dous
sujeitos que se criaram com Gonçalo de Mattos, ambos de instincto
capaz para uma informação, e entre elles achei a contradicção, que
póde servir de exemplo a quem se informa: um affirma com
juramento que era poeta natural, o outro jurando nega que tal fosse,
dizendo que elle nem o Padre Nosso era capaz de repetir. A este
seguem muitos, e nenhum áquelle: mas o primeiro chamado
Christovão Rodrigues diz que em sua adolescencia lhe dera o
seguinte mote:

Com que, porque, para que.

Defendia-se o Gonçalo temeroso de uma maldição condicional de

sua mãe, em respeito da qual não queria pegar na penna para fazer
versos, posto que no animo lhe pulsavam as Musas (tal foi o
escarmento que deixaram ellas naquelles cadaveres da paciencia
lastimosa). Mas como a condição do preceito tinha sua clausula, em
que fundar-se uma heresia graciosa, respondeu importunado:
«Pegae vós na penna, porque a maldição de minha mãe parece que
não me prohibe fazer versos, mas sim pegar na penna para elles.»
Repetiu-me então esta decima, que tanto ella como a resposta, si
são verdadeiras, vem a ser uns relampagos da esphera do fogo:

GLOSA.

Disse Clori que me amava

Para o intento que tem,
O qual não disse a ninguem,
Nem o porque declarava:
Eu então lhe perguntava
Com que genero de fé!
Suspensa a dama se vê;
Como nada respondeu,
Não pude saber o seu
Com que, porque, para que.

Persuado a crer o caso pelas suas circumstancias, e muito mais

quando vejo aqui umas reliquias mais separadas d’aquelle humor,
ou ramas menos fortes do enxerto do doutor Pedro de Mattos seu
tio, onde não ha resposta sem equivoco sem substancia do genero
mais nobre.
Foi o doutor Gregorio de Mattos de boa estatura, secco do corpo,
membros delicados, poucos cabellos e crespos, testa espaçosa,
sobrancelhas arqueadas, olhos garços, nariz aguilenho, bocca
pequena e engraçada, barba sem demasia, e no tracto cortezão.
Trajava commummente seu collete de pelles de ambar, volta de fina
renda, e era finalmente um composto de perfeições como poeta
portuguez, que são Esopos os de outras nações. Tinha phantasia
natural no passeio, e quando algumas vezes por recreação sulcava
os quietos mares da Bahia a remo compassado, com tão bizarra
confiança interpunha os oculos, examinando as janellas da sua
cidade, que muitos curiosos iam de proposito a vê-lo.
 Fiz tirar d’elle a presente cópia, por um antigo pintor, que foi seu
familiar, e conferindo-a com as memorias que d’elle têm algumas
pessoas antigas, tenho-a por mui conforme a seu original. Naquelle
tempo era pouco versado o uso das cabelleiras, e elle a trajava: mas
pareceu-me copia-lo sem ella, porque os homens de talento devem
patentear-nos as officinas capitaes que o produzem para informação
dos judiciosos.
SATYRICAS
 OBRAS POETICAS
DE
GREGORIO DE MATTOS GUERRA

AOS VICIOS
TERCETOS

Eu sou aquelle que os passados annos

Cantei na minha lyra maldizente
Torpezas do Brazil, vicios e enganos.

E bem que os descantei bastantemente,

Canto segunda vez na mesma lyra
O mesmo assumpto em plectro differente.

Já sinto que me inflamma e que me inspira

Thalia, que anjo é da minha guarda
Des’ que Apollo mandou que me assistira.

Arda Bayona, e todo o mundo arda,

Que a quem de profissão falta á verdade
Nunca a dominga das verdades tarda.

Nenhum tempo exceptua a christandade

Ao pobre pegureiro do Parnaso
Para fallar em sua liberdade.

A narração ha de egualar ao caso,

E si talvez ao caso não eguala,
Não tenho por poeta o que é Pegaso.

De que póde servir calar quem cala?

Nunca se ha de fallar o que se sente?!
Sempre se ha de sentir o que se falla.
Qual homem póde haver tão paciente,
Que, vendo o triste estado da Bahia,
Não chore, não suspire e não lamente?

Isto faz a discreta phantasia:

Discorre em um e outro desconcerto,
Condemna o roubo, increpa a hypocrisia.

O nescio, o ignorante, o inexperto,

Que não elege o bom, nem mau reprova,
Por tudo passa deslumbrado e incerto.

E quando vê talvez na doce trova

Louvado o bem, e o mal vituperado,
A tudo faz focinho, e nada approva.

Diz logo prudentaço e repousado:

—Fulano é um satyrico, é um louco,
De lingua má, de coração damnado.

Nescio, si d’isso entendes nada ou pouco,

Como mofas com riso e algazarras
Musas, que estimo ter, quando as invoco.

Si souberas fallar, tambem falláras,

Tambem satyrisáras, si souberas,
E si fôras poeta, poetisáras.

A ignorancia dos homens d’estas eras

Sizudos faz ser uns, outros prudentes,
Que a mudez canoniza bestas feras.

Ha bons, por não poder ser insolentes,

Outros ha comedidos de medrosos,
Não mordem outros não—por não ter dentes.
Quantos ha que os telhados têm vidrosos,
E deixam de atirar sua pedrada,
De sua mesma telha receiosos?

Uma só natureza nos foi dada;

Não creou Deus os naturaes diversos;
Um só Adão creou, e esse de nada.

Todos somos ruins, todos perversos,

Só nos distingue o vicio e a virtude,
De que uns são comensaes, outros adversos.

Quem maior a tiver, do que eu ter pude,

Esse só me censure, esse me note,
Calem-se os mais, chiton, e haja saude.
 BENZE-SE
O P. DE VARIAS ACÇÕES QUE OBSERVAVA NA SUA PATRIA

D’estes que campam no mundo

Sem ter engenho profundo,
E, entre gabos dos amigos,
Os vemos em papafigos
Sem tempestade, nem vento:
Anjo bento!

De quem com lettras secretas

Tudo o que alcança é por tretas,
Baculejando sem pejo,
Por matar o seu desejo,
Desde a manhãa té à tarde:
Deus me guarde!

Do que passeia farfante,

Muito presado de amante,
Por fóra luvas, galões,
Insignias, armas, bastões.
Por dentro pão bolorento:
Anjo bento!

D’estes beatos fingidos,

Cabisbaixos, encolhidos,
Por dentro fataes maganos,
Sendo nas caras uns Janos,
Que fazem do vicio alarde:
Deus me guarde!
 Que vejamos teso andar
Quem mal sabe engatinhar,
Muito inteiro e presumido,
Ficando o outro abatido
Com maior merecimento:
Anjo bento!

D’estes avaros mofinos,

Que põem na meza pepinos,
De toda a iguaria isenta,
Com seu limão e pimenta,
Porque diz que queima e arde:
Deus me guarde!

Que pregue um douto sermão

Um alarve, um asneirão;
E que esgrima em demasia
Quem nunca lá na Sophia[1]
Soube pôr um argumento:
Anjo bento!

[1] Rua principal de Coimbra: allusão á Universidade.

D’esse sancto emmascarado,

Que falla do meu peccado,
E se tem por Sancto Antonio,
Mas em luctas com o demonio
Se mostra sempre cobarde:
Deus me guarde!
Que atropellando a justiça,
Só com virtude postiça,
Se premeie o delinquente,
Castigando o innocente
Por um leve pensamento:
Anjo bento!
 REPROVAÇÕES

Si sois homem valoroso,

Dizem que sois temerario,
Si valente—espadanchim,
E atrevido si esforçado.
Si resoluto—arrogante,
Si pacifico, sois fraco,
Si precatado—medroso,
E si o não sois—confiado.
Si usaes justiça, um Herodes,
Si favoravel, sois brando,
Si condemnaes, sois injusto,
Si absolveis, estaes peitado.
Si vos dão, sois um covarde,
E si daes sois deshumano,
Si vos rendeis, sois traidor,
Si rendeis—afortunado.
Si sois plebeu, sois humilde,
Soberbo, si sois fidalgo,
Si sois segundo sois pobre,
E tolo si sois morgado.
Si galeaes, sois fachada,
E si não—não sois bizarro,
Si vestis bem, sois gram moda,
Si mal vestis sois um trapo.
Si comeis muito, guloso,
E faminto, si sois parco,
Si comeis bem, regalão,
E si mal, nunca sois farto.
Si não soffreis, imprudente,
Si soffreis sois um coitado,
Si perdoaes, sois bom homem,
E si não sois—um tyranno.
Si brioso, tendes fumos,
 E si não, sois homem baixo,
Si sois serio, descortez,
Si cortez, afidalgado.
Si defendeis, sois amigo,
Si o não fazeis sois contrario,
Si sois amigo, suspeito,
Si o não sois—affeiçoado.
Si obraes mal, sois ignorante,
Si bem obraes, foi acaso,
Si não servis, sois isento,
E si servis, sois criado.
Si virtuoso, fingido,
E hypocrita si beato,
Si zeloso, impertinente,
E si não, sois um patrano.
Si sois sisudo, intractavel,
Si sois devoto, sois falso,
Pertinaz, si deffendente,
Si arguinte, porfiado.
Si discreto, prevenido,
E si não, sois insensato,
Si sois modesto, sois simples,
E si o não sois, sois um diabo.
Si sois gracioso, sois fatuo,
E si não sois, um marmanjo,
Si sois agudo tresledes,
E si o não sois, sois um asno.
Si não compondes, sois nescio,
Si escreveis, sois censurado,
Si fazeis versos sois louco,
E si o não fazeis, sois parvo.
Si previsto, feiticeiro,
E si não, desmazelado,
Si verdadeiro, bom homem,
Muito humilde, si sois lhano.
Si robusto, sois grosseiro,
Si delicado, sois brando,
 Si descansado, ocioso,
Si para pouco, sois tranco.
Si sois gordo, sois balôfo,
Sois phthisico, si sois magro,
Si pequeno, sois anão,
E gigante, si sois alto.
Si sois nobre, sois pelão,
E si official, sois baixo,
Si solteiro—extravagante,
Si noivo sois namorado.
Si corado, figadal,
Descorado, si sois alvo,
Si grande nariz—judeu,
Si trigueiro, sois mulato.
Si liberal sois perdido,
E si o não sois, sois escasso,
Si sois prodigo, vicioso,
E avarento, si poupado.
Si não despendeis—mesquinho,
Si despendeis, sois mui largo,
Si não gastaes—miseravel,
Si gastaes—esperdiçado.
Si honesto sois, não sois homem,
... si sois casto,
Si não namoraes, .....
Si o fazeis, sois estragado.
Si não luzis, não sois gente,
Si luzis, sois mui presado,
Si pedis, sois pobretão,
E si não, fazeis Calvarios.
Si andaes devagar—mimoso,
Si depressa sois cavallo,
Mal encarado, si feio,
Si gentil—affeminado.
Si fallaes muito, palreiro,
Si fallaes pouco, sois tardo,
Si em pé, não tendes assento,
 Preguiçoso, si assentado.
E assim não póde viver
Neste Brazil infestado,
Segundo o que vos refiro
Quem não seja reprovado.
 VERDADES

Ouve, ó amigo João,

Esta verdade que canto,
Si a verdade causa espanto
Esta causa admiração:
É certo, sem remissão,
E contra isto não ha nada,
Que a outra verdade usada
Com rebuços, mais enganos,
É verdade de maganos,
Mas esta é de gente honrada.

Domingos e dias sanctos

Nos manda a egreja guardar,
Nos mais dias trabalhar:
As mulheres trazem mantos:
Os doutos estão nos cantos,
Os ignorantes na praça,
Os cachorros vão á caça,
Os gatos furtam as ceias,
Os barbeiros rasgam veias
E as padeiras fazem massa.

Os homens fazem a guerra,

E as mulheres fazem renda,
Os tolos não têm emenda,
Os capos cavam a terra:
O bezerro sem mãe berra
Batem bandeiras alferes,
Os pobres buscam haveres,
Os peixes nadam no mar,
As purgas fazem purgar,
E os franciscanos colhéres.
Os cavallos comem herva,
Os despidos andam nús,
Come o Gentio cajús,
Os Tapuyas são caterva:
Não dorme de noite a cerva,
Os macacos fazem momos,
Os escripturarios tomos,
Os namorados passeiam,
As fragonas zombeteiam,
E as limas todas têm gomos.

Todos os ferrões têm ponta,

A agua do mar é salgada,
O hóspede logo enfada,
Todo o algarismo é conta:
A nau sem vela não monta,
O badalo dá no sino,
Chorar muito é desatino,
Muito comer enche a pança,
Bum-bum é agua em criança
E ter em pé, pino-pino.

Os caranguejos têm pernas,

Tocado o tambor faz bulha,
O arrelá desempulha,
O navio tem cavernas;
O fogo accende as lucernas
Os ... fedem á ..,
Quem degenera não herda,
O carvão todo é de lenha,
É só de lã a estamenha,
E a cabelleira tem cerda.
As hervas são todas folhas,
As larangeiras dão fructas,
Mulheres damas são .. ..,
Uma talha são dez polhas:
As botijas levam rolhas,
Toda a neve é branca e fria,
A irmãa de mãe é tia,
É o bronze todo duro,
Onde não ha luz, é escuro,
Quando não é noite é dia.

O sol e o fogo são quentes,

A chuva aonde cae molha,
Quem não tem vista não olha,
Ossos na boca são dentes:
É affronta dizer—mentes!
É ave grande a gallinha,
O cabello cae com tinha,
Quem é rouco tem catarrho,
Carregado canta o carro,
Mulher de rei é rainha.

Não ha barba sem cabello,

A arêa toda é de grãos,
Toca-se a harpa com as mãos,
É animal o camello:
Nenhuma calva tem pello,
Os ovos saem pelo ..
É marisco o sururú,
Todo o feijão é legume,
Coze-se o comer ao lume,
É abobora o gerumú.
Todo o unguento é mézinha
Não tem banha o bacalhau,
Papas ralas é mingau,
Trigo moido é farinha:
Couza alheia não é minha,
Não ha escada sem degraus,
Os picaros são maraus,
Tem aduellas a pipa,
Umbigo é ponto de tripa,
Sempre é loio o rei de paus.

Primeiro foi frango o gallo,

Palangana é prato fundo,
É redondo todo o mundo,
As luvas não fazem callo:
Tem quatro pés o cavallo,
Nunca mija o papagaio,
O chouriço grande é paio,
Não sabe ler a guariba,
Quem tem carcunda tem giba,
Antes de junho está maio.

Todo o chapeu é sombreiro,

As arvores são de pau,
Tudo o que não presta é mau,
E faz a barba o barbeiro:
O.. detraz é trazeiro,
É nervo a penna de pato,
Filho de parda, é mulato,
Mulheres todas são femeas,
Duas em um ventre são gemeas,
No pé se calça o sapato.