Density Functional Theory (DFT) has been, for over three decades, the standard computational method for predicting materials properties from first principles. Traditionally, DFT has flourished within a “calculation-first”, system-by-system framework. Over time, however, this perspective has been complemented by data-oriented, high-throughput strategies, enabling systematic exploration. This paradigm shift has led to the development of extended, standardised materials databases and, more recently, has fostered the integration into computational materials science of Artificial Intelligence techniques, that allow the identification of patterns and trends hardly detectable on a case-by-case basis. The present research advances materials characterisation by combining the power and accuracy of DFT – embedded where appropriate into automated workflows – with the scalability of high-throughput frameworks. Two families of ferroic systems (displaying at least one emergent property related to magnetic, electric, or mechanical order) are selected as areas of interest, reflecting the intertwining of the theoretical and experimental research lines behind this project. The first part of the work addresses the rapidly developing field of two-dimensional (2D) magnetism. For decades, the existence of long-range magnetic order in strictly 2D systems was considered unlikely on fundamental thermodynamic grounds. This view changed in the late 2010s, when experiments reported stable magnetism in monolayers of FePS₃, CrI₃ and CrGeTe₃. Since then, many additional compounds have been identified, exhibiting a rich variety of magnetic textures, with potential relevance for spintronic technologies. A central task in understanding these materials is the evaluation of exchange parameters, which collectively determine the nature of the magnetic ordering. These parameters are commonly extracted from DFT using approaches such as the “four-state method”, which maps total energies of selected spin configurations onto a Heisenberg-like model Hamiltonian. This technique, however, is typically laborious and multi-step, since each parameter in the model requires four separate DFT calculations, performed in as many independent simulation cells, that also vary depending on the parameter. To overcome this limitation, I designed AMaRaNTA (Automating Magnetic paRAmeters iN a Tensorial Approach), a Python-based computational workflow that automates the four-state method for 2D magnets. Starting solely from a crystal structure file, AMaRaNTA automatically generates the required simulation cells, submits and monitors calculations on high-performance computing facilities, retrieves total energies, and ultimately extracts exchange parameters. In its current implementation, the code returns the nearest-neighbour exchange tensor, along with scalar parameters for second- and third-nearest-neighbour interactions and single-ion anisotropy. These form a minimal yet comprehensive set to capture the mechanisms stabilising diverse magnetic textures in 2D materials. The functionality of AMaRaNTA was validated on a compact dataset of around 30 compounds. The workflow successfully reproduces expected features of well-established 2D magnets, including chromium trihalides (CrX₃, X = Cl, Br, I) and transition-metal dihalides (NiX₂, VX₂, MnX₂). It also captures the unusually strong antiferromagnetic interactions reported for VPS₃ (at the nearest-neighbour level), as well as for NiPS₃ and NiPSe₃ (at the third-nearest-neighbour level). Beyond this, the screening also predicts previously unreported features in selected systems, including antiferromagnetism in NiF₄Tl₂, sizeable Kitaev-like magnetism in MnBi₂Te₄, and antisymmetric Dzyaloshinskii–Moriya interactions in VF₆ and VAgP₂Se₆. Altogether, AMaRaNTA provides a robust, systematic and reproducible strategy for extracting exchange parameters in 2D magnets, with clear potential for a large-scale, high-throughput exploration. The second half of the thesis is devoted to piezoelectric materials. These systems, in which electric and mechanical degrees of freedom are strongly coupled, have a long history of industrial and technological applications. Today, most such devices rely on lead-based compounds, whose toxicity, however, raises significant human health concerns; research is therefore increasingly directed towards the development of lead-free alternative materials. A prominent example is the sodium–potassium niobate solid solution K₁₋ₓNaₓNbO₃ (KNN), a random alloy with the perovskite ABO₃ structure. KNN has attracted significant attention over the past two decades, during which a substantial body of experimental work has provided a general understanding of its composition–temperature phase diagram. By contrast, DFT studies of this system remain limited in number and scope, due to the presence of several competing phases and to the challenges of modelling chemical disorder in random alloys, which require large supercells to accurately capture the multifaceted local environment. My contributions originated as theoretical support to the ongoing activity of in-house experimental collaborators. In this context, I first addressed the experimentally observed tendency of KNN to segregate into Na-rich and K-rich domains. I investigated this behaviour via DFT simulations of idealised, tetragonal KNN systems, thereby disentangling chemical substitution from lattice relaxation effects. Analysis of total energies revealed that the preference for regions of unbalanced composition already emerges at the first-principles level, albeit with a very small associated energy gain. I then turned to a deeper exploration of how KNN properties evolve with composition. After preliminary validation on the sodium niobate end member, I focused on a specific KNN phase, of monoclinic symmetry and intermediate structural complexity, and examined its behaviour on the Na-rich side of the phase diagram. I unveiled how this phase tends to “breathe” with increasing Na content: the lattice vectors in the xz plane alternately open and close, causing the angle β between them to slightly oscillate around 90°. Accordingly, the polarisation follows this structural evolution by progressively oscillating around its reference direction. Analysis of the piezoelectric tensors reveals an overall enhancement of the response at intermediate compositions, corresponding to maximal chemical disorder; the same regime also favours polarisation rotation towards the non-polar ŷ axis. These results are in reasonable agreement with the limited available DFT literature; the calculated d₃₃ piezoelectric coefficient also shows moderate resemblance to our in-house experimental data, within a specific compositional window. Finally, repeating the energetic stability analysis within the more realistic monoclinic framework, I found that the energy gain associated with decomposition towards the end members increases by an order of magnitude with respect to the idealised tetragonal case; this strengthens the earlier conclusion regarding the intrinsic preference of KNN for non-equimolar compositions. Beyond complementing experimental observations, my investigation contributes to the more ambitious goal of establishing a foundation for a systematic, large-scale DFT survey of KNN across phases and compositions.
La Teoria del Funzionale della Densità (Density Functional Theory, DFT) rappresenta da oltre tre decenni il metodo computazionale standard per prevedere le proprietà dei materiali da principi primi. Tradizionalmente, la DFT si è sviluppata all’interno di un approccio “calculation-first”, basato sull’analisi di sistemi specifici caso per caso. Con il tempo, tuttavia, questa prospettiva è stata affiancata da strategie “high-throughput” orientate ai dati, che consentono un’esplorazione sistematica dello spazio dei materiali. Questo cambiamento di paradigma ha portato allo sviluppo di estese banche dati di materiali standardizzate e, più recentemente, ha favorito l’integrazione nella scienza dei materiali computazionale di tecniche di Intelligenza Artificiale, capaci di individuare pattern e tendenze difficilmente rilevabili attraverso analisi caso per caso. La presente ricerca contribuisce al progresso della caratterizzazione dei materiali combinando la potenza e l’accuratezza della DFT – integrata, ove opportuno, in workflow automatizzati – con la scalabilità dei framework high-throughput. Come ambiti di studio vengono considerate due famiglie di sistemi ferroici (caratterizzati dalla presenza di almeno una proprietà emergente legata a un ordine magnetico, elettrico o meccanico), che riflettono l’intreccio tra le linee di ricerca teoriche e sperimentali alla base di questo progetto. La prima parte del lavoro affronta il campo, in rapido sviluppo, del magnetismo bidimensionale (2D). Per decenni l’esistenza di ordine magnetico a lungo raggio in sistemi strettamente bidimensionali è stata considerata improbabile per motivi fondamentali di natura termodinamica. Questa visione è cambiata alla fine degli anni 2010, con l’evidenza sperimentale di magnetismo stabile in monolayer di FePS₃, CrI₃ e CrGeTe₃. Da allora sono stati identificati molti altri composti, che mostrano una grande varietà di texture magnetiche e potenziali applicazioni nelle tecnologie spintroniche. Un compito chiave per la comprensione di questi materiali consiste nella valutazione dei parametri di scambio, che nel loro insieme determinano la natura dell’ordine magnetico. Tali parametri vengono comunemente estratti dalla DFT mediante approcci come il “four-state method”, che mappa le energie totali di configurazioni di spin selezionate su un’Hamiltoniana di tipo Heisenberg. Questa tecnica risulta tuttavia tipicamente laboriosa e articolata in più passaggi, poiché ogni parametro del modello richiede quattro calcoli DFT distinti, eseguiti in altrettante celle di simulazione indipendenti, che variano inoltre a seconda del parametro considerato. Per superare questa limitazione ho progettato AMaRaNTA (Automating Magnetic paRAmeters iN a Tensorial Approach), un workflow computazionale scritto in Python che automatizza il four-state method per magneti bidimensionali. A partire esclusivamente da un file contenente la struttura cristallina, AMaRaNTA genera automaticamente le celle di simulazione necessarie, invia e monitora i calcoli su infrastrutture di calcolo ad alte prestazioni, recupera le energie totali e infine estrae i parametri di scambio. Nell’implementazione attuale, il codice restituisce il tensore di scambio tra primi vicini, insieme a parametri scalari per le interazioni tra secondi e terzi vicini e per la single-ion anisotropy. Questo insieme costituisce un set minimo ma completo per descrivere i meccanismi che stabilizzano diverse texture magnetiche nei materiali bidimensionali. Il funzionamento di AMaRaNTA è stato validato su un dataset compatto di circa 30 composti. Il workflow riproduce con successo caratteristiche attese di magneti 2D ben noti, inclusi i tri-alogenuri di cromo (CrX₃, X = Cl, Br, I) e i di-alogenuri di metalli di transizione (NiX₂, VX₂, MnX₂). Inoltre, cattura le interazioni antiferromagnetiche insolitamente forti riportate per VPS₃ (al livello dei primi vicini), così come per NiPS₃ e NiPSe₃ (al livello dei terzi vicini). Oltre a ciò, lo screening effettuato predice proprietà finora non riportate in alcuni sistemi selezionati, tra cui antiferromagnetismo in NiF₄Tl₂, magnetismo di tipo Kitaev di entità significativa in MnBi₂Te₄ e interazioni antisimmetriche di tipo Dzyaloshinskii–Moriya in VF₄ e VAgP₂Se₆. Nel complesso, AMaRaNTA fornisce una strategia robusta, sistematica e riproducibile per l’estrazione dei parametri di scambio nei magneti bidimensionali, con un chiaro potenziale per esplorazioni high-throughput su larga scala. La seconda metà della tesi è dedicata ai materiali piezoelettrici. In questi sistemi, nei quali i gradi di libertà elettrici e meccanici sono fortemente accoppiati, esiste una lunga tradizione di applicazioni industriali e tecnologiche. Attualmente la maggior parte dei dispositivi si basa su composti contenenti piombo; la loro tossicità solleva tuttavia rilevanti problemi di salute pubblica, motivo per cui la ricerca si sta orientando sempre più verso lo sviluppo di materiali alternativi privi di piombo. Un esempio di particolare interesse è la soluzione solida K₁₋ₓNaₓNbO₃ (niobato di sodio potassio, o KNN), una lega disordinata con la struttura della perovskite ABO₃. Negli ultimi due decenni questo sistema ha attirato grande attenzione e un ampio corpus di studi sperimentali ha fornito una comprensione generale del suo diagramma di fase composizione–temperatura. Al contrario, gli studi basati su DFT rimangono relativamente limitati per numero e portata, a causa della presenza di diverse fasi in competizione e delle difficoltà associate alla modellizzazione del disordine chimico nelle leghe di questo tipo, che richiede supercelle di grandi dimensioni per descrivere accuratamente il complesso ambiente locale. Il mio contributo nasce come supporto teorico all’attività portata avanti da collaboratori sperimentali all'interno del gruppo di ricerca. In questo contesto, ho innanzitutto analizzato la tendenza, osservata sperimentalmente, del KNN a segregare in domini ricchi di sodio e domini ricchi di potassio. Questo comportamento è stato studiato mediante simulazioni DFT di sistemi KNN idealizzati di simmetria tetragonale, separando gli effetti della sostituzione chimica da quelli del rilassamento reticolare. L’analisi delle energie totali mostra che la preferenza per regioni a composizione sbilanciata emerge già a livello di principi primi, sebbene con un guadagno energetico associato molto ridotto. Successivamente, ho approfondito l’evoluzione delle proprietà del KNN al variare della composizione. Dopo una validazione preliminare sul caso limite del niobato di sodio, mi sono concentrato su una specifica fase del KNN, di simmetria monoclina e complessità strutturale intermedia, analizzandone il comportamento nella regione ricca di sodio del diagramma di fase. I risultati mostrano che questa fase tende a “respirare” con l’aumentare del contenuto di sodio: i vettori reticolari nel piano xz si aprono e si chiudono alternativamente, causando una lieve oscillazione dell’angolo β attorno a 90°. Di conseguenza, anche la polarizzazione segue questa evoluzione strutturale, oscillando progressivamente attorno alla sua direzione di riferimento. L’analisi dei tensori piezoelettrici rivela un aumento complessivo della risposta a composizioni intermedie, corrispondenti al massimo grado di disordine chimico; lo stesso regime favorisce inoltre la rotazione della polarizzazione verso l’asse non polare ŷ. Questi risultati sono in ragionevole accordo con la limitata letteratura DFT disponibile; anche il coefficiente piezoelettrico d₃₃ calcolato mostra una moderata somiglianza con i dati sperimentali ottenuti dai nostri collaboratori, all’interno di una specifica finestra composizionale. Infine, ripetendo l’analisi di stabilità energetica nel contesto più realistico della struttura monoclina, emerge che il guadagno energetico associato alla decomposizione verso i casi limite aumenta di circa un ordine di grandezza rispetto al caso tetragonale idealizzato. Questo risultato rafforza la conclusione precedente secondo cui il KNN mostra una preferenza intrinseca per composizioni non equimolari. Oltre a fornire un supporto teorico alle osservazioni sperimentali, questa indagine contribuisce all’obiettivo più ambizioso di costruire le basi per un’indagine sistematica su larga scala, basata su DFT, del sistema KNN al variare delle fasi strutturali e della composizione.
A combined ab initio and high-throughput approach to ferroic materials exploration
Orlando, Federico
2025/2026
Abstract
Density Functional Theory (DFT) has been, for over three decades, the standard computational method for predicting materials properties from first principles. Traditionally, DFT has flourished within a “calculation-first”, system-by-system framework. Over time, however, this perspective has been complemented by data-oriented, high-throughput strategies, enabling systematic exploration. This paradigm shift has led to the development of extended, standardised materials databases and, more recently, has fostered the integration into computational materials science of Artificial Intelligence techniques, that allow the identification of patterns and trends hardly detectable on a case-by-case basis. The present research advances materials characterisation by combining the power and accuracy of DFT – embedded where appropriate into automated workflows – with the scalability of high-throughput frameworks. Two families of ferroic systems (displaying at least one emergent property related to magnetic, electric, or mechanical order) are selected as areas of interest, reflecting the intertwining of the theoretical and experimental research lines behind this project. The first part of the work addresses the rapidly developing field of two-dimensional (2D) magnetism. For decades, the existence of long-range magnetic order in strictly 2D systems was considered unlikely on fundamental thermodynamic grounds. This view changed in the late 2010s, when experiments reported stable magnetism in monolayers of FePS₃, CrI₃ and CrGeTe₃. Since then, many additional compounds have been identified, exhibiting a rich variety of magnetic textures, with potential relevance for spintronic technologies. A central task in understanding these materials is the evaluation of exchange parameters, which collectively determine the nature of the magnetic ordering. These parameters are commonly extracted from DFT using approaches such as the “four-state method”, which maps total energies of selected spin configurations onto a Heisenberg-like model Hamiltonian. This technique, however, is typically laborious and multi-step, since each parameter in the model requires four separate DFT calculations, performed in as many independent simulation cells, that also vary depending on the parameter. To overcome this limitation, I designed AMaRaNTA (Automating Magnetic paRAmeters iN a Tensorial Approach), a Python-based computational workflow that automates the four-state method for 2D magnets. Starting solely from a crystal structure file, AMaRaNTA automatically generates the required simulation cells, submits and monitors calculations on high-performance computing facilities, retrieves total energies, and ultimately extracts exchange parameters. In its current implementation, the code returns the nearest-neighbour exchange tensor, along with scalar parameters for second- and third-nearest-neighbour interactions and single-ion anisotropy. These form a minimal yet comprehensive set to capture the mechanisms stabilising diverse magnetic textures in 2D materials. The functionality of AMaRaNTA was validated on a compact dataset of around 30 compounds. The workflow successfully reproduces expected features of well-established 2D magnets, including chromium trihalides (CrX₃, X = Cl, Br, I) and transition-metal dihalides (NiX₂, VX₂, MnX₂). It also captures the unusually strong antiferromagnetic interactions reported for VPS₃ (at the nearest-neighbour level), as well as for NiPS₃ and NiPSe₃ (at the third-nearest-neighbour level). Beyond this, the screening also predicts previously unreported features in selected systems, including antiferromagnetism in NiF₄Tl₂, sizeable Kitaev-like magnetism in MnBi₂Te₄, and antisymmetric Dzyaloshinskii–Moriya interactions in VF₆ and VAgP₂Se₆. Altogether, AMaRaNTA provides a robust, systematic and reproducible strategy for extracting exchange parameters in 2D magnets, with clear potential for a large-scale, high-throughput exploration. The second half of the thesis is devoted to piezoelectric materials. These systems, in which electric and mechanical degrees of freedom are strongly coupled, have a long history of industrial and technological applications. Today, most such devices rely on lead-based compounds, whose toxicity, however, raises significant human health concerns; research is therefore increasingly directed towards the development of lead-free alternative materials. A prominent example is the sodium–potassium niobate solid solution K₁₋ₓNaₓNbO₃ (KNN), a random alloy with the perovskite ABO₃ structure. KNN has attracted significant attention over the past two decades, during which a substantial body of experimental work has provided a general understanding of its composition–temperature phase diagram. By contrast, DFT studies of this system remain limited in number and scope, due to the presence of several competing phases and to the challenges of modelling chemical disorder in random alloys, which require large supercells to accurately capture the multifaceted local environment. My contributions originated as theoretical support to the ongoing activity of in-house experimental collaborators. In this context, I first addressed the experimentally observed tendency of KNN to segregate into Na-rich and K-rich domains. I investigated this behaviour via DFT simulations of idealised, tetragonal KNN systems, thereby disentangling chemical substitution from lattice relaxation effects. Analysis of total energies revealed that the preference for regions of unbalanced composition already emerges at the first-principles level, albeit with a very small associated energy gain. I then turned to a deeper exploration of how KNN properties evolve with composition. After preliminary validation on the sodium niobate end member, I focused on a specific KNN phase, of monoclinic symmetry and intermediate structural complexity, and examined its behaviour on the Na-rich side of the phase diagram. I unveiled how this phase tends to “breathe” with increasing Na content: the lattice vectors in the xz plane alternately open and close, causing the angle β between them to slightly oscillate around 90°. Accordingly, the polarisation follows this structural evolution by progressively oscillating around its reference direction. Analysis of the piezoelectric tensors reveals an overall enhancement of the response at intermediate compositions, corresponding to maximal chemical disorder; the same regime also favours polarisation rotation towards the non-polar ŷ axis. These results are in reasonable agreement with the limited available DFT literature; the calculated d₃₃ piezoelectric coefficient also shows moderate resemblance to our in-house experimental data, within a specific compositional window. Finally, repeating the energetic stability analysis within the more realistic monoclinic framework, I found that the energy gain associated with decomposition towards the end members increases by an order of magnitude with respect to the idealised tetragonal case; this strengthens the earlier conclusion regarding the intrinsic preference of KNN for non-equimolar compositions. Beyond complementing experimental observations, my investigation contributes to the more ambitious goal of establishing a foundation for a systematic, large-scale DFT survey of KNN across phases and compositions.| File | Dimensione | Formato | |
|---|---|---|---|
|
2026_04_Orlando.pdf
accessibile in internet per tutti
Descrizione: Tesi
Dimensione
49.67 MB
Formato
Adobe PDF
|
49.67 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/256437