i  sapori  dell’arcobalenovedere  colori  e  colorare  il  cibo  con  il  computer

Stefano  BaroniScuola  Internazionale  Superiore  di  Studi  Avanza9

Trieste

breve  conferenza  tenuta  al  Friuli  Future  Forum,  Udine,  27  novembre  2013

il  sapore  dell’arcobaleno

il  sapore  dell’arcobaleno

☛ mercato  globale  di  1.45  miliardi  di  dollari  nel  2009

il  sapore  dell’arcobaleno

☛ mercato  globale  di  1.45  miliardi  di  dollari  nel  2009

☛ per  un  totale  di  50,000  tonnellate  /  anno

il  sapore  dell’arcobaleno

☛ mercato  globale  di  1.45  miliardi  di  dollari  nel  2009

☛ per  un  totale  di  50,000  tonnellate  /  anno

alcolici5%

bibite28%

cibo67%

il  sapore  dell’arcobaleno

☛ mercato  globale  di  1.45  miliardi  di  dollari  nel  2009

☛ per  un  totale  di  50,000  tonnellate  /  anno

alcolici5%

bibite28%

cibo67%

Altri18%

Cina8%

Giappone10%

USA28%

Europa36%

il  sapore  dell’arcobaleno

☛ mercato  globale  di  1.45  miliardi  di  dollari  nel  2009

☛ per  un  totale  di  50,000  tonnellate  /  anno

alcolici5%

bibite28%

cibo67%

Altri18%

Cina8%

Giappone10%

USA28%

Europa36%

☛ mercato  dei  coloran9  naturali  cresciuto  del  35%  nel  quinquennio  2005-­‐2009

“To help businesses discover, develop, and deploy new materials twice as fast, we’re launching what we call the Materials Genome Initiative.

The invention of silicon circuits and lithium ion batteries made computers and iPods and iPads possible, but it took years to get those technologies from the drawing board to the market place. We can do it faster.”

-President Obama (6/11)

“To help businesses discover, develop, and deploy new materials twice as fast, we’re launching what we call the Materials Genome Initiative.

The invention of silicon circuits and lithium ion batteries made computers and iPods and iPads possible, but it took years to get those technologies from the drawing board to the market place. We can do it faster.”

-President Obama (6/11)

$  100  M  requested  in  2012  

“To help businesses discover, develop, and deploy new materials twice as fast, we’re launching what we call the Materials Genome Initiative.

The invention of silicon circuits and lithium ion batteries made computers and iPods and iPads possible, but it took years to get those technologies from the drawing board to the market place. We can do it faster.”

-President Obama (6/11)

NATURE MATERIALS | VOL 12 | MARCH 2013 | www.nature.com/naturematerials 191

Every technology is intimately related to a particular materials set. The steam engines that powered the industrial revolution in the eighteenth century were made of steel and, information

and communication technologies are underpinned by silicon. Once a material is chosen for a given technology, it gets locked with it because of the investments associated with establishing large-scale production lines. This means that changing the materials set in an established technology is a rare event and must be considered as a revolution. Moreover, the initial choice of a material is abso-lutely crucial for the long-lasting success of a technological sector. Importantly, recent times have seen a surge of new technological niches, each one of them potentially looking for a different mate-rials set. Thus, the pressure on the development of new materials is becoming formidable. These should score on many counts. They should be tailored on the specific property that the technology is based on, they often should be compatible with other technologies, should not contain toxic elements, and, if needed in large quanti-ties, should be made of cheap raw materials. As such, searching for materials is a multi-dimensional problem where many boxes should be ticked at the same time.

Although the demand for materials is endlessly growing, experi-mental discovery is bound by high costs and time-consuming procedures of synthesis. Is there another way? Indeed, this is the burgeoning area of computational materials science called ‘high-throughput’ (HT) computational materials design. It is based on the marriage between computational quantum-mechanical–ther-modynamic approaches1,2 and a multitude of techniques rooted in database construction and intelligent data mining3. The concept is simple yet powerful: create a large database containing the cal-culated thermodynamic and electronic properties of existing and hypothetical materials, and then intelligently interrogate the data-base in the search of materials with the desired properties. Clearly, the entire construct should be validated by reality, namely the exist-ing materials must be predicted correctly and the hypothetical ones should eventually be made. Such a reality check feeds back to the theory to construct better databases and increase predictive power.

The high-throughput highway to computational materials designStefano Curtarolo1,2*, Gus L. W. Hart2,3, Marco Buongiorno Nardelli2,4,5, Natalio Mingo2,6, Stefano Sanvito2,7 and Ohad Levy1,2,8

High-throughput computational materials design is an emerging area of materials science. By combining advanced thermo-dynamic and electronic-structure methods with intelligent data mining and database construction, and exploiting the power of current supercomputer architectures, scientists generate, manage and analyse enormous data repositories for the discovery of novel materials. In this Review we provide a current snapshot of this rapidly evolving field, and highlight the challenges and opportunities that lie ahead.

The HT experimental approach was pioneered over a hundred years ago by Edison4 and Ciamician5, but with the advent of effi-cient and accurate theoretical tools and inexpensive computers, its computational counterpart has become a viable path for tackling materials design. Thus, in the past decade computational HT materi-als research has emerged3,6–16 following the impetus of experimental HT approaches17–19. In the literature, HT materials research is often confused with the combinatorial evaluation of materials properties. Although a few attempts have been made to clearly define the two concepts20–22, the distinction is not yet rigorous. Here we define HT as the throughput of data that is way too high to be produced or ana-lysed by the researcher’s direct intervention, and must therefore be performed automatically: HT implies an automatic flow from ideas to results. The confusion of HT with combinatorial approaches is thus resolved. The latter, in fact, specifies how the degrees of free-dom are investigated, whereas HT strictly defines the overwhelming and automatic flow of the investigations.

The practical implementation of computational HT is highly non-trivial. The method is employed in three strictly connected steps: (i) virtual materials growth: thermodynamic and electronic structure calculations of materials3,23; (ii) rational materials storage: systematic storage of the information in database repositories24,25; (iii) materials characterization and selection: data analysis aimed at selecting novel materials or gaining new physical insights15,19,26.

High-throughput is often known for the large databases it gen-erates (for example, the AFLOWLIB.org consortium24 and the Materials Project25). Here we posit that all three HT stages are highly necessary, but that the last one is the most challenging and impor-tant. In fact, it is the step that allows one to extract the information and, as such, it requires a deep understanding of the physical prob-lem at hand. The intelligent search of a database is performed by means of ‘descriptors’. These are empirical quantities, not necessarily observables, connecting the calculated microscopic parameters (for example, formation and defect energies, atomic environments, band structure, density of states or magnetic moments) to macroscopic properties of the materials (for example, mobility, susceptibility or

1Department of Mechanical Engineering and Materials Science, and Department of Physics, Duke University, Durham, North Carolina 27708, USA, 2Center for Materials Genomics, Duke University, Durham, North Carolina 27708, USA, 3Department of Physics and Astronomy, Brigham Young University, Provo, Utah 84602, USA, 4Department of Physics and Department of Chemistry, University of North Texas, Denton, Texas 76203, 5Computer Science and Mathematics Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831, USA, 6LITEN, CEA-Grenoble, 17 rue des Martyrs, 38054 Grenoble Cedex 9, France, 7School of Physics and CRANN, Trinity College, Dublin 2, Ireland, 8Department of Physics, NRCN, PO Box 9001, Beer-Sheva 84190, Israel. *e-mail: stefano@duke.edu

REVIEW ARTICLEPUBLISHED ONLINE: 20 FEBRUARY 2013"|"DOI: 10.1038/NMAT3568

© 2013 Macmillan Publishers Limited. All rights reserved

NATURE MATERIALS | VOL 12 | MARCH 2013 | www.nature.com/naturematerials 191

Every technology is intimately related to a particular materials set. The steam engines that powered the industrial revolution in the eighteenth century were made of steel and, information

and communication technologies are underpinned by silicon. Once a material is chosen for a given technology, it gets locked with it because of the investments associated with establishing large-scale production lines. This means that changing the materials set in an established technology is a rare event and must be considered as a revolution. Moreover, the initial choice of a material is abso-lutely crucial for the long-lasting success of a technological sector. Importantly, recent times have seen a surge of new technological niches, each one of them potentially looking for a different mate-rials set. Thus, the pressure on the development of new materials is becoming formidable. These should score on many counts. They should be tailored on the specific property that the technology is based on, they often should be compatible with other technologies, should not contain toxic elements, and, if needed in large quanti-ties, should be made of cheap raw materials. As such, searching for materials is a multi-dimensional problem where many boxes should be ticked at the same time.

Although the demand for materials is endlessly growing, experi-mental discovery is bound by high costs and time-consuming procedures of synthesis. Is there another way? Indeed, this is the burgeoning area of computational materials science called ‘high-throughput’ (HT) computational materials design. It is based on the marriage between computational quantum-mechanical–ther-modynamic approaches1,2 and a multitude of techniques rooted in database construction and intelligent data mining3. The concept is simple yet powerful: create a large database containing the cal-culated thermodynamic and electronic properties of existing and hypothetical materials, and then intelligently interrogate the data-base in the search of materials with the desired properties. Clearly, the entire construct should be validated by reality, namely the exist-ing materials must be predicted correctly and the hypothetical ones should eventually be made. Such a reality check feeds back to the theory to construct better databases and increase predictive power.

The high-throughput highway to computational materials designStefano Curtarolo1,2*, Gus L. W. Hart2,3, Marco Buongiorno Nardelli2,4,5, Natalio Mingo2,6, Stefano Sanvito2,7 and Ohad Levy1,2,8

High-throughput computational materials design is an emerging area of materials science. By combining advanced thermo-dynamic and electronic-structure methods with intelligent data mining and database construction, and exploiting the power of current supercomputer architectures, scientists generate, manage and analyse enormous data repositories for the discovery of novel materials. In this Review we provide a current snapshot of this rapidly evolving field, and highlight the challenges and opportunities that lie ahead.

The HT experimental approach was pioneered over a hundred years ago by Edison4 and Ciamician5, but with the advent of effi-cient and accurate theoretical tools and inexpensive computers, its computational counterpart has become a viable path for tackling materials design. Thus, in the past decade computational HT materi-als research has emerged3,6–16 following the impetus of experimental HT approaches17–19. In the literature, HT materials research is often confused with the combinatorial evaluation of materials properties. Although a few attempts have been made to clearly define the two concepts20–22, the distinction is not yet rigorous. Here we define HT as the throughput of data that is way too high to be produced or ana-lysed by the researcher’s direct intervention, and must therefore be performed automatically: HT implies an automatic flow from ideas to results. The confusion of HT with combinatorial approaches is thus resolved. The latter, in fact, specifies how the degrees of free-dom are investigated, whereas HT strictly defines the overwhelming and automatic flow of the investigations.

The practical implementation of computational HT is highly non-trivial. The method is employed in three strictly connected steps: (i) virtual materials growth: thermodynamic and electronic structure calculations of materials3,23; (ii) rational materials storage: systematic storage of the information in database repositories24,25; (iii) materials characterization and selection: data analysis aimed at selecting novel materials or gaining new physical insights15,19,26.

High-throughput is often known for the large databases it gen-erates (for example, the AFLOWLIB.org consortium24 and the Materials Project25). Here we posit that all three HT stages are highly necessary, but that the last one is the most challenging and impor-tant. In fact, it is the step that allows one to extract the information and, as such, it requires a deep understanding of the physical prob-lem at hand. The intelligent search of a database is performed by means of ‘descriptors’. These are empirical quantities, not necessarily observables, connecting the calculated microscopic parameters (for example, formation and defect energies, atomic environments, band structure, density of states or magnetic moments) to macroscopic properties of the materials (for example, mobility, susceptibility or

1Department of Mechanical Engineering and Materials Science, and Department of Physics, Duke University, Durham, North Carolina 27708, USA, 2Center for Materials Genomics, Duke University, Durham, North Carolina 27708, USA, 3Department of Physics and Astronomy, Brigham Young University, Provo, Utah 84602, USA, 4Department of Physics and Department of Chemistry, University of North Texas, Denton, Texas 76203, 5Computer Science and Mathematics Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831, USA, 6LITEN, CEA-Grenoble, 17 rue des Martyrs, 38054 Grenoble Cedex 9, France, 7School of Physics and CRANN, Trinity College, Dublin 2, Ireland, 8Department of Physics, NRCN, PO Box 9001, Beer-Sheva 84190, Israel. *e-mail: stefano@duke.edu

REVIEW ARTICLEPUBLISHED ONLINE: 20 FEBRUARY 2013"|"DOI: 10.1038/NMAT3568

© 2013 Macmillan Publishers Limited. All rights reserved

NATURE MATERIALS | VOL 12 | MARCH 2013 | www.nature.com/naturematerials 191

Every technology is intimately related to a particular materials set. The steam engines that powered the industrial revolution in the eighteenth century were made of steel and, information

and communication technologies are underpinned by silicon. Once a material is chosen for a given technology, it gets locked with it because of the investments associated with establishing large-scale production lines. This means that changing the materials set in an established technology is a rare event and must be considered as a revolution. Moreover, the initial choice of a material is abso-lutely crucial for the long-lasting success of a technological sector. Importantly, recent times have seen a surge of new technological niches, each one of them potentially looking for a different mate-rials set. Thus, the pressure on the development of new materials is becoming formidable. These should score on many counts. They should be tailored on the specific property that the technology is based on, they often should be compatible with other technologies, should not contain toxic elements, and, if needed in large quanti-ties, should be made of cheap raw materials. As such, searching for materials is a multi-dimensional problem where many boxes should be ticked at the same time.

Although the demand for materials is endlessly growing, experi-mental discovery is bound by high costs and time-consuming procedures of synthesis. Is there another way? Indeed, this is the burgeoning area of computational materials science called ‘high-throughput’ (HT) computational materials design. It is based on the marriage between computational quantum-mechanical–ther-modynamic approaches1,2 and a multitude of techniques rooted in database construction and intelligent data mining3. The concept is simple yet powerful: create a large database containing the cal-culated thermodynamic and electronic properties of existing and hypothetical materials, and then intelligently interrogate the data-base in the search of materials with the desired properties. Clearly, the entire construct should be validated by reality, namely the exist-ing materials must be predicted correctly and the hypothetical ones should eventually be made. Such a reality check feeds back to the theory to construct better databases and increase predictive power.

The high-throughput highway to computational materials designStefano Curtarolo1,2*, Gus L. W. Hart2,3, Marco Buongiorno Nardelli2,4,5, Natalio Mingo2,6, Stefano Sanvito2,7 and Ohad Levy1,2,8

High-throughput computational materials design is an emerging area of materials science. By combining advanced thermo-dynamic and electronic-structure methods with intelligent data mining and database construction, and exploiting the power of current supercomputer architectures, scientists generate, manage and analyse enormous data repositories for the discovery of novel materials. In this Review we provide a current snapshot of this rapidly evolving field, and highlight the challenges and opportunities that lie ahead.

The HT experimental approach was pioneered over a hundred years ago by Edison4 and Ciamician5, but with the advent of effi-cient and accurate theoretical tools and inexpensive computers, its computational counterpart has become a viable path for tackling materials design. Thus, in the past decade computational HT materi-als research has emerged3,6–16 following the impetus of experimental HT approaches17–19. In the literature, HT materials research is often confused with the combinatorial evaluation of materials properties. Although a few attempts have been made to clearly define the two concepts20–22, the distinction is not yet rigorous. Here we define HT as the throughput of data that is way too high to be produced or ana-lysed by the researcher’s direct intervention, and must therefore be performed automatically: HT implies an automatic flow from ideas to results. The confusion of HT with combinatorial approaches is thus resolved. The latter, in fact, specifies how the degrees of free-dom are investigated, whereas HT strictly defines the overwhelming and automatic flow of the investigations.

The practical implementation of computational HT is highly non-trivial. The method is employed in three strictly connected steps: (i) virtual materials growth: thermodynamic and electronic structure calculations of materials3,23; (ii) rational materials storage: systematic storage of the information in database repositories24,25; (iii) materials characterization and selection: data analysis aimed at selecting novel materials or gaining new physical insights15,19,26.

High-throughput is often known for the large databases it gen-erates (for example, the AFLOWLIB.org consortium24 and the Materials Project25). Here we posit that all three HT stages are highly necessary, but that the last one is the most challenging and impor-tant. In fact, it is the step that allows one to extract the information and, as such, it requires a deep understanding of the physical prob-lem at hand. The intelligent search of a database is performed by means of ‘descriptors’. These are empirical quantities, not necessarily observables, connecting the calculated microscopic parameters (for example, formation and defect energies, atomic environments, band structure, density of states or magnetic moments) to macroscopic properties of the materials (for example, mobility, susceptibility or

1Department of Mechanical Engineering and Materials Science, and Department of Physics, Duke University, Durham, North Carolina 27708, USA, 2Center for Materials Genomics, Duke University, Durham, North Carolina 27708, USA, 3Department of Physics and Astronomy, Brigham Young University, Provo, Utah 84602, USA, 4Department of Physics and Department of Chemistry, University of North Texas, Denton, Texas 76203, 5Computer Science and Mathematics Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831, USA, 6LITEN, CEA-Grenoble, 17 rue des Martyrs, 38054 Grenoble Cedex 9, France, 7School of Physics and CRANN, Trinity College, Dublin 2, Ireland, 8Department of Physics, NRCN, PO Box 9001, Beer-Sheva 84190, Israel. *e-mail: stefano@duke.edu

REVIEW ARTICLEPUBLISHED ONLINE: 20 FEBRUARY 2013"|"DOI: 10.1038/NMAT3568

© 2013 Macmillan Publishers Limited. All rights reserved

Every technology is intimately related to a particular materials set. The steam engines that powered the industrial revolution in the eighteenth century were made of steel and, information

and communication technologies are underpinned by silicon. Once a material is chosen for a given technology, it gets locked with it because of the investments associated with establishing large-scale production lines. This means that changing the materials set in an established technology is a rare event and must be considered as a revolution. Moreover, the initial choice of a material is abso-lutely crucial for the long-lasting success of a technological sector. Importantly, recent times have seen a surge of new technological niches, each one of them potentially looking for a different mate-rials set. Thus, the pressure on the development of new materials is becoming formidable. These should score on many counts. They should be tailored on the specific property that the technology is based on, they often should be compatible with other technologies, should not contain toxic elements, and, if needed in large quanti-ties, should be made of cheap raw materials. As such, searching for materials is a multi-dimensional problem where many boxes should be ticked at the same time.

Although the demand for materials is endlessly growing, experi-

NATURE MATERIALS | VOL 12 | MARCH 2013 | www.nature.com/naturematerials 191

Every technology is intimately related to a particular materials set. The steam engines that powered the industrial revolution in the eighteenth century were made of steel and, information

and communication technologies are underpinned by silicon. Once a material is chosen for a given technology, it gets locked with it because of the investments associated with establishing large-scale production lines. This means that changing the materials set in an established technology is a rare event and must be considered as a revolution. Moreover, the initial choice of a material is abso-lutely crucial for the long-lasting success of a technological sector. Importantly, recent times have seen a surge of new technological niches, each one of them potentially looking for a different mate-rials set. Thus, the pressure on the development of new materials is becoming formidable. These should score on many counts. They should be tailored on the specific property that the technology is based on, they often should be compatible with other technologies, should not contain toxic elements, and, if needed in large quanti-ties, should be made of cheap raw materials. As such, searching for materials is a multi-dimensional problem where many boxes should be ticked at the same time.

Although the demand for materials is endlessly growing, experi-mental discovery is bound by high costs and time-consuming procedures of synthesis. Is there another way? Indeed, this is the burgeoning area of computational materials science called ‘high-throughput’ (HT) computational materials design. It is based on the marriage between computational quantum-mechanical–ther-modynamic approaches1,2 and a multitude of techniques rooted in database construction and intelligent data mining3. The concept is simple yet powerful: create a large database containing the cal-culated thermodynamic and electronic properties of existing and hypothetical materials, and then intelligently interrogate the data-base in the search of materials with the desired properties. Clearly, the entire construct should be validated by reality, namely the exist-ing materials must be predicted correctly and the hypothetical ones should eventually be made. Such a reality check feeds back to the theory to construct better databases and increase predictive power.

The high-throughput highway to computational materials designStefano Curtarolo1,2*, Gus L. W. Hart2,3, Marco Buongiorno Nardelli2,4,5, Natalio Mingo2,6, Stefano Sanvito2,7 and Ohad Levy1,2,8

High-throughput computational materials design is an emerging area of materials science. By combining advanced thermo-dynamic and electronic-structure methods with intelligent data mining and database construction, and exploiting the power of current supercomputer architectures, scientists generate, manage and analyse enormous data repositories for the discovery of novel materials. In this Review we provide a current snapshot of this rapidly evolving field, and highlight the challenges and opportunities that lie ahead.

The HT experimental approach was pioneered over a hundred years ago by Edison4 and Ciamician5, but with the advent of effi-cient and accurate theoretical tools and inexpensive computers, its computational counterpart has become a viable path for tackling materials design. Thus, in the past decade computational HT materi-als research has emerged3,6–16 following the impetus of experimental HT approaches17–19. In the literature, HT materials research is often confused with the combinatorial evaluation of materials properties. Although a few attempts have been made to clearly define the two concepts20–22, the distinction is not yet rigorous. Here we define HT as the throughput of data that is way too high to be produced or ana-lysed by the researcher’s direct intervention, and must therefore be performed automatically: HT implies an automatic flow from ideas to results. The confusion of HT with combinatorial approaches is thus resolved. The latter, in fact, specifies how the degrees of free-dom are investigated, whereas HT strictly defines the overwhelming and automatic flow of the investigations.

The practical implementation of computational HT is highly non-trivial. The method is employed in three strictly connected steps: (i) virtual materials growth: thermodynamic and electronic structure calculations of materials3,23; (ii) rational materials storage: systematic storage of the information in database repositories24,25; (iii) materials characterization and selection: data analysis aimed at selecting novel materials or gaining new physical insights15,19,26.

High-throughput is often known for the large databases it gen-erates (for example, the AFLOWLIB.org consortium24 and the Materials Project25). Here we posit that all three HT stages are highly necessary, but that the last one is the most challenging and impor-tant. In fact, it is the step that allows one to extract the information and, as such, it requires a deep understanding of the physical prob-lem at hand. The intelligent search of a database is performed by means of ‘descriptors’. These are empirical quantities, not necessarily observables, connecting the calculated microscopic parameters (for example, formation and defect energies, atomic environments, band structure, density of states or magnetic moments) to macroscopic properties of the materials (for example, mobility, susceptibility or

1Department of Mechanical Engineering and Materials Science, and Department of Physics, Duke University, Durham, North Carolina 27708, USA, 2Center for Materials Genomics, Duke University, Durham, North Carolina 27708, USA, 3Department of Physics and Astronomy, Brigham Young University, Provo, Utah 84602, USA, 4Department of Physics and Department of Chemistry, University of North Texas, Denton, Texas 76203, 5Computer Science and Mathematics Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831, USA, 6LITEN, CEA-Grenoble, 17 rue des Martyrs, 38054 Grenoble Cedex 9, France, 7School of Physics and CRANN, Trinity College, Dublin 2, Ireland, 8Department of Physics, NRCN, PO Box 9001, Beer-Sheva 84190, Israel. *e-mail: stefano@duke.edu

REVIEW ARTICLEPUBLISHED ONLINE: 20 FEBRUARY 2013"|"DOI: 10.1038/NMAT3568

© 2013 Macmillan Publishers Limited. All rights reserved

Every technology is intimately related to a particular materials set. The steam engines that powered the industrial revolution in the eighteenth century were made of steel and, information

and communication technologies are underpinned by silicon. Once a material is chosen for a given technology, it gets locked with it because of the investments associated with establishing large-scale production lines. This means that changing the materials set in an established technology is a rare event and must be considered as a revolution. Moreover, the initial choice of a material is abso-lutely crucial for the long-lasting success of a technological sector. Importantly, recent times have seen a surge of new technological niches, each one of them potentially looking for a different mate-rials set. Thus, the pressure on the development of new materials is becoming formidable. These should score on many counts. They should be tailored on the specific property that the technology is based on, they often should be compatible with other technologies, should not contain toxic elements, and, if needed in large quanti-ties, should be made of cheap raw materials. As such, searching for materials is a multi-dimensional problem where many boxes should be ticked at the same time.

Although the demand for materials is endlessly growing, experi-

NATURE MATERIALS | VOL 12 | MARCH 2013 | www.nature.com/naturematerials 191

Every technology is intimately related to a particular materials set. The steam engines that powered the industrial revolution in the eighteenth century were made of steel and, information

and communication technologies are underpinned by silicon. Once a material is chosen for a given technology, it gets locked with it because of the investments associated with establishing large-scale production lines. This means that changing the materials set in an established technology is a rare event and must be considered as a revolution. Moreover, the initial choice of a material is abso-lutely crucial for the long-lasting success of a technological sector. Importantly, recent times have seen a surge of new technological niches, each one of them potentially looking for a different mate-rials set. Thus, the pressure on the development of new materials is becoming formidable. These should score on many counts. They should be tailored on the specific property that the technology is based on, they often should be compatible with other technologies, should not contain toxic elements, and, if needed in large quanti-ties, should be made of cheap raw materials. As such, searching for materials is a multi-dimensional problem where many boxes should be ticked at the same time.

Although the demand for materials is endlessly growing, experi-mental discovery is bound by high costs and time-consuming procedures of synthesis. Is there another way? Indeed, this is the burgeoning area of computational materials science called ‘high-throughput’ (HT) computational materials design. It is based on the marriage between computational quantum-mechanical–ther-modynamic approaches1,2 and a multitude of techniques rooted in database construction and intelligent data mining3. The concept is simple yet powerful: create a large database containing the cal-culated thermodynamic and electronic properties of existing and hypothetical materials, and then intelligently interrogate the data-base in the search of materials with the desired properties. Clearly, the entire construct should be validated by reality, namely the exist-ing materials must be predicted correctly and the hypothetical ones should eventually be made. Such a reality check feeds back to the theory to construct better databases and increase predictive power.

The high-throughput highway to computational materials designStefano Curtarolo1,2*, Gus L. W. Hart2,3, Marco Buongiorno Nardelli2,4,5, Natalio Mingo2,6, Stefano Sanvito2,7 and Ohad Levy1,2,8

High-throughput computational materials design is an emerging area of materials science. By combining advanced thermo-dynamic and electronic-structure methods with intelligent data mining and database construction, and exploiting the power of current supercomputer architectures, scientists generate, manage and analyse enormous data repositories for the discovery of novel materials. In this Review we provide a current snapshot of this rapidly evolving field, and highlight the challenges and opportunities that lie ahead.

The HT experimental approach was pioneered over a hundred years ago by Edison4 and Ciamician5, but with the advent of effi-cient and accurate theoretical tools and inexpensive computers, its computational counterpart has become a viable path for tackling materials design. Thus, in the past decade computational HT materi-als research has emerged3,6–16 following the impetus of experimental HT approaches17–19. In the literature, HT materials research is often confused with the combinatorial evaluation of materials properties. Although a few attempts have been made to clearly define the two concepts20–22, the distinction is not yet rigorous. Here we define HT as the throughput of data that is way too high to be produced or ana-lysed by the researcher’s direct intervention, and must therefore be performed automatically: HT implies an automatic flow from ideas to results. The confusion of HT with combinatorial approaches is thus resolved. The latter, in fact, specifies how the degrees of free-dom are investigated, whereas HT strictly defines the overwhelming and automatic flow of the investigations.

The practical implementation of computational HT is highly non-trivial. The method is employed in three strictly connected steps: (i) virtual materials growth: thermodynamic and electronic structure calculations of materials3,23; (ii) rational materials storage: systematic storage of the information in database repositories24,25; (iii) materials characterization and selection: data analysis aimed at selecting novel materials or gaining new physical insights15,19,26.

High-throughput is often known for the large databases it gen-erates (for example, the AFLOWLIB.org consortium24 and the Materials Project25). Here we posit that all three HT stages are highly necessary, but that the last one is the most challenging and impor-tant. In fact, it is the step that allows one to extract the information and, as such, it requires a deep understanding of the physical prob-lem at hand. The intelligent search of a database is performed by means of ‘descriptors’. These are empirical quantities, not necessarily observables, connecting the calculated microscopic parameters (for example, formation and defect energies, atomic environments, band structure, density of states or magnetic moments) to macroscopic properties of the materials (for example, mobility, susceptibility or

1Department of Mechanical Engineering and Materials Science, and Department of Physics, Duke University, Durham, North Carolina 27708, USA, 2Center for Materials Genomics, Duke University, Durham, North Carolina 27708, USA, 3Department of Physics and Astronomy, Brigham Young University, Provo, Utah 84602, USA, 4Department of Physics and Department of Chemistry, University of North Texas, Denton, Texas 76203, 5Computer Science and Mathematics Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831, USA, 6LITEN, CEA-Grenoble, 17 rue des Martyrs, 38054 Grenoble Cedex 9, France, 7School of Physics and CRANN, Trinity College, Dublin 2, Ireland, 8Department of Physics, NRCN, PO Box 9001, Beer-Sheva 84190, Israel. *e-mail: stefano@duke.edu

REVIEW ARTICLEPUBLISHED ONLINE: 20 FEBRUARY 2013"|"DOI: 10.1038/NMAT3568

© 2013 Macmillan Publishers Limited. All rights reserved

Every technology is intimately related to a particular materials set. The steam engines that powered the industrial revolution in the eighteenth century were made of steel and, information

and communication technologies are underpinned by silicon. Once a material is chosen for a given technology, it gets locked with it because of the investments associated with establishing large-scale production lines. This means that changing the materials set in an established technology is a rare event and must be considered as a revolution. Moreover, the initial choice of a material is abso-lutely crucial for the long-lasting success of a technological sector. Importantly, recent times have seen a surge of new technological niches, each one of them potentially looking for a different mate-rials set. Thus, the pressure on the development of new materials is becoming formidable. These should score on many counts. They should be tailored on the specific property that the technology is based on, they often should be compatible with other technologies, should not contain toxic elements, and, if needed in large quanti-ties, should be made of cheap raw materials. As such, searching for materials is a multi-dimensional problem where many boxes should be ticked at the same time.

Although the demand for materials is endlessly growing, experi-

NATURE MATERIALS | VOL 12 | MARCH 2013 | www.nature.com/naturematerials 191

Every technology is intimately related to a particular materials set. The steam engines that powered the industrial revolution in the eighteenth century were made of steel and, information

and communication technologies are underpinned by silicon. Once a material is chosen for a given technology, it gets locked with it because of the investments associated with establishing large-scale production lines. This means that changing the materials set in an established technology is a rare event and must be considered as a revolution. Moreover, the initial choice of a material is abso-lutely crucial for the long-lasting success of a technological sector. Importantly, recent times have seen a surge of new technological niches, each one of them potentially looking for a different mate-rials set. Thus, the pressure on the development of new materials is becoming formidable. These should score on many counts. They should be tailored on the specific property that the technology is based on, they often should be compatible with other technologies, should not contain toxic elements, and, if needed in large quanti-ties, should be made of cheap raw materials. As such, searching for materials is a multi-dimensional problem where many boxes should be ticked at the same time.

Although the demand for materials is endlessly growing, experi-mental discovery is bound by high costs and time-consuming procedures of synthesis. Is there another way? Indeed, this is the burgeoning area of computational materials science called ‘high-throughput’ (HT) computational materials design. It is based on the marriage between computational quantum-mechanical–ther-modynamic approaches1,2 and a multitude of techniques rooted in database construction and intelligent data mining3. The concept is simple yet powerful: create a large database containing the cal-culated thermodynamic and electronic properties of existing and hypothetical materials, and then intelligently interrogate the data-base in the search of materials with the desired properties. Clearly, the entire construct should be validated by reality, namely the exist-ing materials must be predicted correctly and the hypothetical ones should eventually be made. Such a reality check feeds back to the theory to construct better databases and increase predictive power.

The high-throughput highway to computational materials designStefano Curtarolo1,2*, Gus L. W. Hart2,3, Marco Buongiorno Nardelli2,4,5, Natalio Mingo2,6, Stefano Sanvito2,7 and Ohad Levy1,2,8

High-throughput computational materials design is an emerging area of materials science. By combining advanced thermo-dynamic and electronic-structure methods with intelligent data mining and database construction, and exploiting the power of current supercomputer architectures, scientists generate, manage and analyse enormous data repositories for the discovery of novel materials. In this Review we provide a current snapshot of this rapidly evolving field, and highlight the challenges and opportunities that lie ahead.

The HT experimental approach was pioneered over a hundred years ago by Edison4 and Ciamician5, but with the advent of effi-cient and accurate theoretical tools and inexpensive computers, its computational counterpart has become a viable path for tackling materials design. Thus, in the past decade computational HT materi-als research has emerged3,6–16 following the impetus of experimental HT approaches17–19. In the literature, HT materials research is often confused with the combinatorial evaluation of materials properties. Although a few attempts have been made to clearly define the two concepts20–22, the distinction is not yet rigorous. Here we define HT as the throughput of data that is way too high to be produced or ana-lysed by the researcher’s direct intervention, and must therefore be performed automatically: HT implies an automatic flow from ideas to results. The confusion of HT with combinatorial approaches is thus resolved. The latter, in fact, specifies how the degrees of free-dom are investigated, whereas HT strictly defines the overwhelming and automatic flow of the investigations.

The practical implementation of computational HT is highly non-trivial. The method is employed in three strictly connected steps: (i) virtual materials growth: thermodynamic and electronic structure calculations of materials3,23; (ii) rational materials storage: systematic storage of the information in database repositories24,25; (iii) materials characterization and selection: data analysis aimed at selecting novel materials or gaining new physical insights15,19,26.

High-throughput is often known for the large databases it gen-erates (for example, the AFLOWLIB.org consortium24 and the Materials Project25). Here we posit that all three HT stages are highly necessary, but that the last one is the most challenging and impor-tant. In fact, it is the step that allows one to extract the information and, as such, it requires a deep understanding of the physical prob-lem at hand. The intelligent search of a database is performed by means of ‘descriptors’. These are empirical quantities, not necessarily observables, connecting the calculated microscopic parameters (for example, formation and defect energies, atomic environments, band structure, density of states or magnetic moments) to macroscopic properties of the materials (for example, mobility, susceptibility or

1Department of Mechanical Engineering and Materials Science, and Department of Physics, Duke University, Durham, North Carolina 27708, USA, 2Center for Materials Genomics, Duke University, Durham, North Carolina 27708, USA, 3Department of Physics and Astronomy, Brigham Young University, Provo, Utah 84602, USA, 4Department of Physics and Department of Chemistry, University of North Texas, Denton, Texas 76203, 5Computer Science and Mathematics Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831, USA, 6LITEN, CEA-Grenoble, 17 rue des Martyrs, 38054 Grenoble Cedex 9, France, 7School of Physics and CRANN, Trinity College, Dublin 2, Ireland, 8Department of Physics, NRCN, PO Box 9001, Beer-Sheva 84190, Israel. *e-mail: stefano@duke.edu

REVIEW ARTICLEPUBLISHED ONLINE: 20 FEBRUARY 2013"|"DOI: 10.1038/NMAT3568

© 2013 Macmillan Publishers Limited. All rights reserved

Every technology is intimately related to a particular materials set. The steam engines that powered the industrial revolution in the eighteenth century were made of steel and, information

and communication technologies are underpinned by silicon. Once a material is chosen for a given technology, it gets locked with it because of the investments associated with establishing large-scale production lines. This means that changing the materials set in an established technology is a rare event and must be considered as a revolution. Moreover, the initial choice of a material is abso-lutely crucial for the long-lasting success of a technological sector. Importantly, recent times have seen a surge of new technological niches, each one of them potentially looking for a different mate-rials set. Thus, the pressure on the development of new materials is becoming formidable. These should score on many counts. They should be tailored on the specific property that the technology is based on, they often should be compatible with other technologies, should not contain toxic elements, and, if needed in large quanti-ties, should be made of cheap raw materials. As such, searching for materials is a multi-dimensional problem where many boxes should be ticked at the same time.

Although the demand for materials is endlessly growing, experi-

coloran9  alimentari

tecnologia  di  nicchia

coloran9  alimentari

tecnologia  di  nicchia

proprietà  specifica:  il  colore

coloran9  alimentari

tecnologia  di  nicchia

proprietà  specifica:  il  colore

✔ valore  di  mercato:  naturalezza

coloran9  alimentari

tecnologia  di  nicchia

proprietà  specifica:  il  colore

✔ valore  di  mercato:  naturalezza

✔ valore  finanziario:  abbondanza  (naturale)

coloran9  alimentari

tecnologia  di  nicchia

proprietà  specifica:  il  colore

✔ valore  di  mercato:  naturalezza

✔ valore  finanziario:  abbondanza  (naturale)

✔ valore  tecnico:  flessibilità  funzionale

coloran9  alimentari

tecnologia  di  nicchia

proprietà  specifica:  il  colore

✔ valore  di  mercato:  naturalezza

✔ valore  finanziario:  abbondanza  (naturale)

✔ valore  tecnico:  flessibilità  funzionale

coloran9  alimentari

=            antocianine

tecnologia  di  nicchia

proprietà  specifica:  il  colore

✔ valore  di  mercato:  naturalezza

✔ valore  finanziario:  abbondanza  (naturale)

✔ valore  tecnico:  flessibilità  funzionale

coloran9  alimentari

=            antocianine

anthocyanins

sugar

chromenyliumphenyl

anthocyanin R1 R2 R3 R7

cyanin −OH −OH −H −OH

peonin −OCH3 −OH −H −OH

rosinin −OH −OH −H −OCH3

malvin −OCH3 −OH −OCH3 −OH

delphinin −OH −OH −OCH3 −OH

pelargonin −H −OH −OH −OH

antho-­‐0 −H −H −H −OH

anthocyanins

sugar

chromenyliumphenyl

anthocyanin R1 R2 R3 R7

cyanin −OH −OH −H −OH

peonin −OCH3 −OH −H −OH

rosinin −OH −OH −H −OCH3

malvin −OCH3 −OH −OCH3 −OH

delphinin −OH −OH −OCH3 −OH

pelargonin −H −OH −OH −OH

antho-­‐0 −H −H −H −OH

anthocyanins

sugar

chromenyliumphenyl

anthocyanin R1 R2 R3 R7

cyanin −OH −OH −H −OH

peonin −OCH3 −OH −H −OH

rosinin −OH −OH −H −OCH3

malvin −OCH3 −OH −OCH3 −OH

delphinin −OH −OH −OCH3 −OH

pelargonin −H −OH −OH −OH

antho-­‐0 −H −H −H −OH

anthocyanins

sugar

chromenyliumphenyl

anthocyanin R1 R2 R3 R7

cyanin −OH −OH −H −OH

peonin −OCH3 −OH −H −OH

rosinin −OH −OH −H −OCH3

malvin −OCH3 −OH −OCH3 −OH

delphinin −OH −OH −OCH3 −OH

pelargonin −H −OH −OH −OH

antho-­‐0 −H −H −H −OH

anthocyanins

sugar

chromenyliumphenyl

anthocyanin R1 R2 R3 R7

cyanin −OH −OH −H −OH

peonin −OCH3 −OH −H −OH

rosinin −OH −OH −H −OCH3

malvin −OCH3 −OH −OCH3 −OH

delphinin −OH −OH −OCH3 −OH

pelargonin −H −OH −OH −OH

antho-­‐0 −H −H −H −OH

anthocyanins

sugar

chromenyliumphenyl

anthocyanin R1 R2 R3 R7

cyanin −OH −OH −H −OH

peonin −OCH3 −OH −H −OH

rosinin −OH −OH −H −OCH3

malvin −OCH3 −OH −OCH3 −OH

delphinin −OH −OH −OCH3 −OH

pelargonin −H −OH −OH −OH

antho-­‐0 −H −H −H −OH

anthocyanins

sugar

chromenyliumphenyl

1.1. ANTHOCYANINS

Figure 1.4: The main four equilibrium forms of anthocyanin existing in aqueous media

[31].

Figure 1.5: The distribution of the di↵erent Malvindin-3-glucoside equilibrium forms

according to pH [19].

are often organic acids, which can esterify the glycosyl units of anthocyanins.

The aromatic ring of the organic acid will then fold to surround the antho-

cyanin, thus a↵ecting the optical properties of the molecule both directly,

and indirectly through the modification of its structure [30, 31]. Due to

5

anthocyanins:  the  role  of  acidity

pH1 131.1. ANTHOCYANINS

Figure 1.4: The main four equilibrium forms of anthocyanin existing in aqueous media[31].

Figure 1.5: The distribution of the di↵erent Malvindin-3-glucoside equilibrium formsaccording to pH [19].

are often organic acids, which can esterify the glycosyl units of anthocyanins.The aromatic ring of the organic acid will then fold to surround the antho-cyanin, thus a↵ecting the optical properties of the molecule both directly,and indirectly through the modification of its structure [30, 31]. Due to

5

anthocyanins:  the  role  of  hydroxyla9on

anthocyanins:  the  role  of  copigmenta9on

anthocyanins:  the  hurdles  towards  a  ra9onal  design

anthocyanins:  the  hurdles  towards  a  ra9onal  design

the  stability  and  color  func9on  of  anthocyanins  are  affected  by  many  and  diverse  factors:

structural  diversity  (phenols,  sugars,  and  acyla9on)  pH  sensi9vityco-­‐pigmenta9on  

anthocyanins:  the  hurdles  towards  a  ra9onal  design

the  stability  and  color  func9on  of  anthocyanins  are  affected  by  many  and  diverse  factors:

structural  diversity  (phenols,  sugars,  and  acyla9on)  pH  sensi9vityco-­‐pigmenta9on  

the  high  reac9vity  of  the  (phenolic)  chromophore  makes  synthesis  extremely  difficult

most  of  research  simply  aims  at  isola9ng  from  natural  sources  (highly  expensive  and  difficult)very  liale  research  is  being  done  in  this  area

anthocyanins:  the  hurdles  towards  a  ra9onal  design

the  stability  and  color  func9on  of  anthocyanins  are  affected  by  many  and  diverse  factors:

structural  diversity  (phenols,  sugars,  and  acyla9on)  pH  sensi9vityco-­‐pigmenta9on  

the  high  reac9vity  of  the  (phenolic)  chromophore  makes  synthesis  extremely  difficult

most  of  research  simply  aims  at  isola9ng  from  natural  sources  (highly  expensive  and  difficult)very  liale  research  is  being  done  in  this  area

very  liale  is  known  on  the  microscopic  mechanisms  that  determine  the  stability  and  the  chroma9c  proper9es  of  anthocyanins  and  the  rela9on  between  structure  an  color

molecular  and  materials  modeling

back  in  the  fibies

molecular  and  materials  modeling

back  in  the  fibies

1962

molecular  and  materials  modeling

back  in  the  fibies

1962

third  millennium

molecular  modeling

back  in  the  fibies

1962

third  millennium

2013

Michael  Levia

Arieh  WarshelMar9n  Karplus

"for  the  development  of  

mul9scale  [computa9onal]

models  for  complex  

chemical  systems"

what  color  is  all  about

?

what  color  is  all  about

what  color  is  all  about

what  color  is  all  about

what  color  is  all  about

anycolor(λ)  =  r(λ)  +  g(λ)  +  b(λ)

450 550 650

reflec9on  vs.  transmission

reflec9on  vs.  transmission

reflec9on  vs.  transmission

CHAPTER 1. INTRODUCTION

the protection of the copigment, the stability of anthoycanins can also beincreased [16]. This is another important aspect of copimentation whichbrings a lot of commercial value to the research of this phenomenon [33].An example of the copigmentation in nature is the bluish purple flowers ofthe Japanese garden iris [34]. Even though fundamental for the color ex-pression, we will not consider explicitly copigmentation in the rest of thisthesis.

1.2 Simulating molecular colors

When a beam of light impinges on the surface of a material, several di↵er-ent processes occur as illustrated in Fig. (1.6). Part of the light is directlyreflected by the surface, while the rest is transmitted into the material. Theamount of reflected light depends on the refractive index of the material, thesmoothness of the surface and the incidence angle ✓. This process gives riseto the so-called surface gloss [35]. The surface gloss under a white lightsource is usually also white, despite the fact that the material itself mayhave other colors. However materials with a strong optical dispersion (i.e.with a refractive index that strongly depends on the wavelength) display acolored gloss, such as metals.

Figure 1.6: Illustration of light interacting with material.

6

what  makes  things  gliaer  the  way  they  do

s9mulus  =  illuminant  ×  trasmission  ×  sensi9vity

what  makes  things  gliaer  the  way  they  do

s9mulus  =  illuminant  ×  trasmission  ×  sensi9vity

S(�)

what  makes  things  gliaer  the  way  they  do

s9mulus  =  illuminant  ×  trasmission  ×  sensi9vity

!

!"#!

$"#!

""#!

%"#!

&"#'()*+,

-./01234(560.7(589:47;+!"#$

S(�)⇥ e��(⇥)x

what  makes  things  gliaer  the  way  they  do

s9mulus  =  illuminant  ×  trasmission  ×  sensi9vity

!

!"#!

$"#!

""#!

%"#!

&"#'()*+,

-./01234(560.7(589:47;+!"#$

T(x,�)

x

absorbing  mediumlight

400 500 600 700

κ(λ)

λ[nm]

S(�)⇥ e��(⇥)x

what  makes  things  gliaer  the  way  they  do

s9mulus  =  illuminant  ×  trasmission  ×  sensi9vity

!

!"#!

$"#!

""#!

%"#!

&"#'()*+,

-./01234(560.7(589:47;+!"#$

S(�)⇥ e��(⇥)x ⇥ rgb(�)400 500 600 700

κ(λ)

λ[nm]

what  makes  things  gliaer  the  way  they  do

s9mulus  =  illuminant  ×  trasmission  ×  sensi9vity

!

!"#!

$"#!

""#!

%"#!

&"#'()*+,

-./01234(560.7(589:47;+!"#$

b rg

rgb

λ[nm]

400 500 600 700

κ(λ)

λ[nm]

RGB(x) =

ZS(�)e��(⇥)xrgb(�)d�

what  makes  things  gliaer  the  way  they  do

!

!"#!

$"#!

""#!

%"#!

&"#'()*+,

-./01234(560.7(589:47;+!"#$

b rg

rgb

λ[nm]

a  puzzle  for  you

a  puzzle  for  you

hint:  the  answer  is  contained  in  one  of  the  previous  slides  

what  computer  modeling  is  all  about

the  saga  of  9me  and  length  scales

10-15 10-12 10-9 10-6 10-3

time [s]

length [m]

10-9

10-6

10-3

nano scale� = 1

� = 0macro scale

the  saga  of  9me  and  length  scales

10-15 10-12 10-9 10-6 10-3

time [s]

length [m]

10-9

10-6

10-3

nano scale� = 1

� = 0macro scale

hic sunt leones

the  saga  of  9me  and  length  scales

10-15 10-12 10-9 10-6 10-3

time [s]

length [m]

10-9

10-6

10-3

nano scale� = 1

� = 0macro scale

hic sunt leones

classical (electro-) dynamics, thermodynamics &

finite elements

kinetic Monte Carlo

electronic structure methods

classical moleculardynamics

classical empirical methods☛ pair potentials☛ force fields☛ shell models

size  vs.  accuracy  of  atomis9c  modeling

quantum many-body methods☛ quantum Monte Carlo☛ MP2, CCSD(T), CI☛ GW, BSE

quantum empirical methods☛ tight-binding☛ embedded atom

quantum self-consistent methods ☛ density Functional Theory☛ Hartree-Fock

size

accuracy

classical empirical methods☛ pair potentials☛ force fields☛ shell models

size  vs.  accuracy  of  atomis9c  modeling

quantum many-body methods☛ quantum Monte Carlo☛ MP2, CCSD(T), CI☛ GW, BSE

quantum empirical methods☛ tight-binding☛ embedded atom

quantum self-consistent methods ☛ density Functional Theory☛ Hartree-Fock

size

accuracy

i�⇥�(r,R; t)⇥t

=�� �2

2M

⇥2

⇥R2� �2

2m

⇥2

⇥r2+ V (r,R)

⇥�(r,R; t)

ab  ini9o  simula9ons

MR̈ = �⇥E(R)⇥R�

� �2

2m

⇥2

⇥r2+ V (r,R)

⇥�(r|R) = E(R)�(r|R)

M≫m:  the  Born-­‐Oppenheimer  approxima9on  

i�⇥�(r,R; t)⇥t

=�� �2

2M

⇥2

⇥R2� �2

2m

⇥2

⇥r2+ V (r,R)

⇥�(r,R; t)

ab  ini9o  simula9ons

from  chemistry  to  color

pelargoninC21H21O10

from  chemistry  to  color

pelargoninC21H21O10

from  chemistry  to  color

pelargoninC21H21O10

HOMO-­‐1

from  chemistry  to  color

pelargoninC21H21O10

3.2. EFFECTS OF SIDE GROUPS

0

-2.5

-2

-1.5

pelargonin peonin malvin

KS o

rbita

l ene

rgy

(ev)

sugarphenyl

chromenylium

A

B

C

A

B

C

A

B

C

L

Figure 3.6: Colored KS energy levels of three representative anthocyanins. Di↵erentcolors indicate the spatial distribution of the orbital (see text).

Anthocyanins Pel Cya Peo Del Mal PetH ! L 1.65 1.50 1.59 1.64 1.51 1.55H ! L + 1 3.17 3.05 3.11 3.17 3.01 3.07

Table 3.1: Energy di↵erence of HOMO to LUMO, and HOMO to LUMO+1 of antho-cyanins [ eV ].

ual energy levels both via the influence on the molecular structure and theelectrostatic e↵ect (see Fig. (3.6)). The energy of orbital A increases withthe number of �OR groups. This is mainly because the �OR group formsan anti-bonding with the A orbital, hence introduces an extra node in theorbital, see Fig. (3.7). For the same reason the energy of orbital C is alsoincreased by the �OR group. The energy of orbital C is more sensitive tothe �OR group, because orbital C is localized on the phenyl ring, thereforeit is more a↵ected by the change on this ring. In the case of pelargonin andpeonin, orbital C has lower energy than both A and B, but in malvin, thereare three �OR groups which lifts the energy of C to be higher than theorbital B.

The oscillator strength of individual transitions is a↵ected by the ap-proximate symmetry of the molecular orbitals involved in them. In orderto explain the change of orbital symmetry, we define an axis of symmetryfor the penyl part of anthocyanin in In Fig. (3.8). In Fig. (3.7) it can beseen that the phenyl ring part of all the LUMO orbitals have approximatelythe even symmetry with respect to this axis, as well as the orbital A andB of pelargonin and malvin.2 The orbital C of pelargonin and malvin has

2 If the molecular orbital rotating along the axis by 180� gives the same wavefunc-

65

energy HOMO-­‐1

HOMO-­‐1LUMO

from  chemistry  to  color

pelargoninC21H21O10

3.2. EFFECTS OF SIDE GROUPS

0

-2.5

-2

-1.5

pelargonin peonin malvin

KS o

rbita

l ene

rgy

(ev)

sugarphenyl

chromenylium

A

B

C

A

B

C

A

B

C

L

Figure 3.6: Colored KS energy levels of three representative anthocyanins. Di↵erentcolors indicate the spatial distribution of the orbital (see text).

Anthocyanins Pel Cya Peo Del Mal PetH ! L 1.65 1.50 1.59 1.64 1.51 1.55H ! L + 1 3.17 3.05 3.11 3.17 3.01 3.07

Table 3.1: Energy di↵erence of HOMO to LUMO, and HOMO to LUMO+1 of antho-cyanins [ eV ].

ual energy levels both via the influence on the molecular structure and theelectrostatic e↵ect (see Fig. (3.6)). The energy of orbital A increases withthe number of �OR groups. This is mainly because the �OR group formsan anti-bonding with the A orbital, hence introduces an extra node in theorbital, see Fig. (3.7). For the same reason the energy of orbital C is alsoincreased by the �OR group. The energy of orbital C is more sensitive tothe �OR group, because orbital C is localized on the phenyl ring, thereforeit is more a↵ected by the change on this ring. In the case of pelargonin andpeonin, orbital C has lower energy than both A and B, but in malvin, thereare three �OR groups which lifts the energy of C to be higher than theorbital B.

The oscillator strength of individual transitions is a↵ected by the ap-proximate symmetry of the molecular orbitals involved in them. In orderto explain the change of orbital symmetry, we define an axis of symmetryfor the penyl part of anthocyanin in In Fig. (3.8). In Fig. (3.7) it can beseen that the phenyl ring part of all the LUMO orbitals have approximatelythe even symmetry with respect to this axis, as well as the orbital A andB of pelargonin and malvin.2 The orbital C of pelargonin and malvin has

2 If the molecular orbital rotating along the axis by 180� gives the same wavefunc-

65

energy

LUMO

HOMO-­‐1

HOMO-­‐1HOMO-­‐4LUMO

from  chemistry  to  color

pelargoninC21H21O10

3.2. EFFECTS OF SIDE GROUPS

0

-2.5

-2

-1.5

pelargonin peonin malvin

KS o

rbita

l ene

rgy

(ev)

sugarphenyl

chromenylium

A

B

C

A

B

C

A

B

C

L

Figure 3.6: Colored KS energy levels of three representative anthocyanins. Di↵erentcolors indicate the spatial distribution of the orbital (see text).

Anthocyanins Pel Cya Peo Del Mal PetH ! L 1.65 1.50 1.59 1.64 1.51 1.55H ! L + 1 3.17 3.05 3.11 3.17 3.01 3.07

Table 3.1: Energy di↵erence of HOMO to LUMO, and HOMO to LUMO+1 of antho-cyanins [ eV ].

ual energy levels both via the influence on the molecular structure and theelectrostatic e↵ect (see Fig. (3.6)). The energy of orbital A increases withthe number of �OR groups. This is mainly because the �OR group formsan anti-bonding with the A orbital, hence introduces an extra node in theorbital, see Fig. (3.7). For the same reason the energy of orbital C is alsoincreased by the �OR group. The energy of orbital C is more sensitive tothe �OR group, because orbital C is localized on the phenyl ring, thereforeit is more a↵ected by the change on this ring. In the case of pelargonin andpeonin, orbital C has lower energy than both A and B, but in malvin, thereare three �OR groups which lifts the energy of C to be higher than theorbital B.

The oscillator strength of individual transitions is a↵ected by the ap-proximate symmetry of the molecular orbitals involved in them. In orderto explain the change of orbital symmetry, we define an axis of symmetryfor the penyl part of anthocyanin in In Fig. (3.8). In Fig. (3.7) it can beseen that the phenyl ring part of all the LUMO orbitals have approximatelythe even symmetry with respect to this axis, as well as the orbital A andB of pelargonin and malvin.2 The orbital C of pelargonin and malvin has

2 If the molecular orbital rotating along the axis by 180� gives the same wavefunc-

65

energy

LUMO

HOMO-­‐1

HOMO-­‐4

HOMO-­‐1HOMO-­‐4LUMO

from  chemistry  to  color

pelargoninC21H21O10

3.2. EFFECTS OF SIDE GROUPS

0

-2.5

-2

-1.5

pelargonin peonin malvin

KS o

rbita

l ene

rgy

(ev)

sugarphenyl

chromenylium

A

B

C

A

B

C

A

B

C

L

Figure 3.6: Colored KS energy levels of three representative anthocyanins. Di↵erentcolors indicate the spatial distribution of the orbital (see text).

Anthocyanins Pel Cya Peo Del Mal PetH ! L 1.65 1.50 1.59 1.64 1.51 1.55H ! L + 1 3.17 3.05 3.11 3.17 3.01 3.07

Table 3.1: Energy di↵erence of HOMO to LUMO, and HOMO to LUMO+1 of antho-cyanins [ eV ].

ual energy levels both via the influence on the molecular structure and theelectrostatic e↵ect (see Fig. (3.6)). The energy of orbital A increases withthe number of �OR groups. This is mainly because the �OR group formsan anti-bonding with the A orbital, hence introduces an extra node in theorbital, see Fig. (3.7). For the same reason the energy of orbital C is alsoincreased by the �OR group. The energy of orbital C is more sensitive tothe �OR group, because orbital C is localized on the phenyl ring, thereforeit is more a↵ected by the change on this ring. In the case of pelargonin andpeonin, orbital C has lower energy than both A and B, but in malvin, thereare three �OR groups which lifts the energy of C to be higher than theorbital B.

The oscillator strength of individual transitions is a↵ected by the ap-proximate symmetry of the molecular orbitals involved in them. In orderto explain the change of orbital symmetry, we define an axis of symmetryfor the penyl part of anthocyanin in In Fig. (3.8). In Fig. (3.7) it can beseen that the phenyl ring part of all the LUMO orbitals have approximatelythe even symmetry with respect to this axis, as well as the orbital A andB of pelargonin and malvin.2 The orbital C of pelargonin and malvin has

2 If the molecular orbital rotating along the axis by 180� gives the same wavefunc-

65

energy

LUMO

HOMO-­‐1

HOMO-­‐4

400 500 600 700

3 2.5 2

Abso

rptio

n

Wavelength (nm)

Energy (ev)

HOMO-­‐1HOMO-­‐4LUMO

from  chemistry  to  color

pelargoninC21H21O10

3.2. EFFECTS OF SIDE GROUPS

0

-2.5

-2

-1.5

pelargonin peonin malvin

KS o

rbita

l ene

rgy

(ev)

sugarphenyl

chromenylium

A

B

C

A

B

C

A

B

C

L

Figure 3.6: Colored KS energy levels of three representative anthocyanins. Di↵erentcolors indicate the spatial distribution of the orbital (see text).

Anthocyanins Pel Cya Peo Del Mal PetH ! L 1.65 1.50 1.59 1.64 1.51 1.55H ! L + 1 3.17 3.05 3.11 3.17 3.01 3.07

Table 3.1: Energy di↵erence of HOMO to LUMO, and HOMO to LUMO+1 of antho-cyanins [ eV ].

ual energy levels both via the influence on the molecular structure and theelectrostatic e↵ect (see Fig. (3.6)). The energy of orbital A increases withthe number of �OR groups. This is mainly because the �OR group formsan anti-bonding with the A orbital, hence introduces an extra node in theorbital, see Fig. (3.7). For the same reason the energy of orbital C is alsoincreased by the �OR group. The energy of orbital C is more sensitive tothe �OR group, because orbital C is localized on the phenyl ring, thereforeit is more a↵ected by the change on this ring. In the case of pelargonin andpeonin, orbital C has lower energy than both A and B, but in malvin, thereare three �OR groups which lifts the energy of C to be higher than theorbital B.

The oscillator strength of individual transitions is a↵ected by the ap-proximate symmetry of the molecular orbitals involved in them. In orderto explain the change of orbital symmetry, we define an axis of symmetryfor the penyl part of anthocyanin in In Fig. (3.8). In Fig. (3.7) it can beseen that the phenyl ring part of all the LUMO orbitals have approximatelythe even symmetry with respect to this axis, as well as the orbital A andB of pelargonin and malvin.2 The orbital C of pelargonin and malvin has

2 If the molecular orbital rotating along the axis by 180� gives the same wavefunc-

65

energy

LUMO

HOMO-­‐1

HOMO-­‐4

400 500 600 700

3 2.5 2

Abso

rptio

n

Wavelength (nm)

Energy (ev)

C55H72MgN4O

chlorofyll  a

400 500 600 700

!

" [nm]

tddftexpt

chlorofyll  a

400 500 600 700

!

" [nm]

tddftexpt

chlorofyll  a

color  and  func9on  of  anthocyanins

cyanidin-­‐3-­‐glucoside

TDDFT  ?

color  and  func9on  of  anthocyanins

cyanidin-­‐3-­‐glucoside

TDDFT  ?TDDFT  :-­‐(

color  and  func9on  of  anthocyanins

cyanidin-­‐3-­‐glucoside

300 400 500 600 700

absorp9o

n

λ  [nm]

tddfptoctopusgaussian

op9cal  effect  of  the  solvent!"#$%&'$

(

400 500 600

)!#&*+

,-.(+/

op9cal  effect  of  the  solvent!"#$%&'$

(

400 500 600

)!#&*+

,-.(+/

C21H21O11Cl@(H2O)95339  atoms938  electrons

op9cal  effect  of  the  solvent!"#$%&'$

(

400 500 600

)!#&*+

,-.(+/

C21H21O11Cl@(H2O)95339  atoms938  electrons

op9cal  effect  of  the  solvent!"#$%&'$

(

400 500 600

)!#&*+

,-.(+/

C21H21O11Cl@(H2O)95339  atoms938  electrons

op9cal  effect  of  the  solvent!"#$%&'$

(

400 500 600

)!#&*+

,-.(+/

400 500 600 700

3 2.5 2

Abso

rptio

n

Wavelength (nm)

Energy (ev)

avg

C21H21O11Cl@(H2O)95339  atoms938  electrons

op9cal  effect  of  the  solvent!"#$%&'$

(

400 500 600

)!#&*+

,-.(+/

400 500 600 700

3 2.5 2

Abso

rptio

n

Wavelength (nm)

Energy (ev)

avg

400 500 600 700

3 2.5 2

Abso

rptio

n

Wavelength (nm)

Energy (ev)

expt

the  MARISA  way  to  molecular  design

the  MARISA  way  to  molecular  design

• Set  up  and  benchmark  a  mul9scale  modeling  framework  for  simula9ng  the  molecular  structure  and  thermal  fluctua9ons  in  realis9c  solva9on  environments.  This  framework  will  be  based  on  advanced  embedding  techniques,  such  as:

• MD  (Molecular  Dynamics)

• QM/MM  (Quantum  Mechanics/Molecular  Mechanics)

• PCM  (Polarizable  Con9nuum  Model).

the  MARISA  way  to  molecular  design

• Set  up  and  benchmark  a  mul9scale  modeling  framework  for  simula9ng  the  molecular  structure  and  thermal  fluctua9ons  in  realis9c  solva9on  environments.  This  framework  will  be  based  on  advanced  embedding  techniques,  such  as:

• MD  (Molecular  Dynamics)

• QM/MM  (Quantum  Mechanics/Molecular  Mechanics)

• PCM  (Polarizable  Con9nuum  Model).

• Benchmark  state-­‐of-­‐the  art  quantum  mechanical  modeling  techniques  against  specific  a  molecular  proper9es  (color)  of  a  specific  class  of  molecules  (anthocyanins).  These  techniques  will  be  mainly  based  on  

• TDDFT  (Time-­‐Dependent  Density-­‐Func9onal  Theory).

the  MARISA  way  to  molecular  design

• Set  up  and  benchmark  a  mul9scale  modeling  framework  for  simula9ng  the  molecular  structure  and  thermal  fluctua9ons  in  realis9c  solva9on  environments.  This  framework  will  be  based  on  advanced  embedding  techniques,  such  as:

• MD  (Molecular  Dynamics)

• QM/MM  (Quantum  Mechanics/Molecular  Mechanics)

• PCM  (Polarizable  Con9nuum  Model).

• Benchmark  state-­‐of-­‐the  art  quantum  mechanical  modeling  techniques  against  specific  a  molecular  proper9es  (color)  of  a  specific  class  of  molecules  (anthocyanins).  These  techniques  will  be  mainly  based  on  

• TDDFT  (Time-­‐Dependent  Density-­‐Func9onal  Theory).

• Develop  approximate  schemes  for  quantum-­‐mechanical  calcula9ons  that  retain  the  accuracy  of  state-­‐of-­‐the-­‐art  techniques,  but  are  tailored  to  an  op9mal  performance  for  the  chroma9c  proper9es  of  anthocyanins.

the  MARISA  way  to  molecular  design

• Set  up  and  benchmark  a  mul9scale  modeling  framework  for  simula9ng  the  molecular  structure  and  thermal  fluctua9ons  in  realis9c  solva9on  environments.  This  framework  will  be  based  on  advanced  embedding  techniques,  such  as:

• MD  (Molecular  Dynamics)

• QM/MM  (Quantum  Mechanics/Molecular  Mechanics)

• PCM  (Polarizable  Con9nuum  Model).

• Benchmark  state-­‐of-­‐the  art  quantum  mechanical  modeling  techniques  against  specific  a  molecular  proper9es  (color)  of  a  specific  class  of  molecules  (anthocyanins).  These  techniques  will  be  mainly  based  on  

• TDDFT  (Time-­‐Dependent  Density-­‐Func9onal  Theory).

• Develop  approximate  schemes  for  quantum-­‐mechanical  calcula9ons  that  retain  the  accuracy  of  state-­‐of-­‐the-­‐art  techniques,  but  are  tailored  to  an  op9mal  performance  for  the  chroma9c  proper9es  of  anthocyanins.

• Use  those  approximate  schemes  for  the  high-­‐throughput  screening  of  large  numbers  of  candidate  anthocyanins  for  a  desired  property  (blue  color)  in  specific  condi9ons  of  temperature,  acidity,  etc.  

Stefano  Baroni,  fisico,  PI

Arrigo  Calzolari,  scienziato  dei  materiali,  consulente  (CNR,  Modena)

la  squadra  MARISA  @SISSA

Iurii  Timrov,  fisico

Alessandro  Biancardi,  chimico  

XiaoChuan  Ge,  fisico,  studente  di  PhD

grazie di esser quabaroni@sissa.it

http://talks.baroni.me