Making Tabular Data FAIR: Applying CSV on the Web (CSVW) for Semantic Interoperability in Computational Science and Engineering

Making Tabular Data FAIR: Applying CSV on the Web (CSVW) for Semantic Interoperability in Computational Science and Engineering Thomas Koprucki joint work with: Robert Finn (Tyndall National Institute, Ireland) Dorothea Iglezakis (Uni Stuttgart) Kickoff Meeting of the GAMM Activity Group on Research Software Engineering and Research Data Management in Mathematics & Mechanics December 05, 2025; TU Braunschweig 1 2024-12-05 Weierstrass Institute for Applied Analysis and Stochastics Interdisciplinary Highlight About me Kickoff Meeting of the GAMM Activity Group on Research Software Engineering and Research Data Management in Mathematics & Mechanics /102 2025-12-05 🎓 Education ▪Diploma in Physics (University of Paderborn) ▪PhD in Mathematics (Freie Universität Berlin) 🏢 Professional Career ▪Since 1997 at Weierstrass Institute for Applied Analysis and Stochastics (WIAS), Berlin ▪Staff Scientist (permanent), Research Group „Partial Differential Equations“ 🔬 Research ▪Mathematical modeling and simulation of nanoand optoelectronic devices ▪Electronic structure and transport in semicond. ▪Numerical methods for semiconductor physics 🛠 RSE & 🗂 RDM ▪Learned RSE from Jürgen Fuhrmann (WIAS) ▪Co-initiator and co-coordinator of the Mathematical Research Data Initiative (MaRDI) ▪FAIR principles for mathematical and simulation-based research Highlight Tabular data or data frames in computational science and engineering Kickoff Meeting of the GAMM Activity Group on Research Software Engineering and Research Data Management in Mathematics & Mechanics /103 2025-12-05 Tabular Data or Data Frames Metadata and FAIR-principles ▪Limited metadata: table schema and data types ▪Missing: units, mapping columns to quantities or quantity kinds ▪Often description in README or annotations in column header Carrier_Density Eu_1.3nm Eu_2.3nm Eu_3.3nm 1.0000E+18 12.6070 14.1550 15.4630 1.0000E+19 12.4410 12.7330 14.5940 1.0000E+20 12.6070 11.1130 12.3680 1 ▪Organization of data in a 2D table of rows and columns ▪Commonly used data scheme for input and output for simulations and data processing ▪Basis for scientific visualizations, plots and figures ▪File formats: CSV, TSV, HDF5, NetCFD ▪Machine-readable, library support ▪Used in data publications accompanying journal papers (e.g. Zenodo) Rows Columns Data Findability Reusability Incomplete metadata has impact on Highlight CSV on the Web (CSVW) Kickoff Meeting of the GAMM Activity Group on Research Software Engineering and Research Data Management in Mathematics & Mechanics /104 Example: https://csvw.org/ Grit bin image: https://commons.wikimedia.org/w/index.php?curid=63203377 2025-12-05 ▪Comma-Separated Values Files (CSV): ITEF RFC, but no schema how to interpret the fields ▪CSV on the Web: W3C standard (2015) to add metadata to CSV data files Wiki Commons, By Rodhullandemu, CC BY-SA 4.0 Semantic annotation Data type Example: location of grit bins in Leeds JSON-LD CSV dialect Highlight CSV on the Web (CSVW) Kickoff Meeting of the GAMM Activity Group on Research Software Engineering and Research Data Management in Mathematics & Mechanics /105 1 https://www.gov.uk/government/publications/recommended-open-standards-for-government/using-metadata-to-describe-csv-data 2 https://cldf.clld.org/ and https://github.com/cldf/csvw 2025-12-05 Aim: machine-readable, interoperable semantic annotation to describe the structure and contents of CSV data files Details and Properties ▪JSON-LD-based metadata file accompanying the CSV file ▪Describes columns, data types, units, and semantic meanings, links to other resources ▪Enables semantic annotations using ontologies such as Dublin Core, Schema.org or custom ontologies like QUDT ▪Supports validation, extensive test suite ▪Tool support (linter, validator, converter) and library support (Python, R), Jupyter notebooks ▪MIME type application/csvm+json ▪Adoption: UK Government since 2020 for OpenData1, Cross-Linguistic Data Formats2 (ERC, MPI) ▪Approach to implement FAIR or Tim Berners-Lee's 5-star principles ▪But no widespread adoption by the research data (NFDI) community so far? Highlight Case study form Computational Materials Science Kickoff Meeting of the GAMM Activity Group on Research Software Engineering and Research Data Management in Mathematics & Mechanics /106 CLDF/CSVW library: GitHub: https://github.com/cldf/csvw PyPI: https://pypi.org/project/csvw 2025-12-05 Background ▪Atomistic simulations of electronic structure of Aluminum Gallium Nitride semiconductors ▪Development of UVC laser diodes for applications in disinfection and life sciences (Leibniz project UVSimTec) ▪Publication in Phys. Rev. Applied PHYSICAL REVIEW APPLIED 24, 044084 (2025) Impact of carrier-density screening on Urbach-tail energies and optical polarization in (Al,Ga)N quantum well systems Robert Finn ,1,*Michael O’Donovan ,2Thomas Koprucki ,2and Stefan Schulz 1,3 1Tyndall National Institute,University College Cork, Cork T12 R5CP, Ireland 2Weierstrass Institute (WIAS), Mohrenstr. 39, 10117 Berlin, Germany 3School of Physics, University College Cork, Cork T12 YN60, Ireland (Received 20 November 2024; revised 12 June 2025; accepted 4 September 2025; published 27 October 2025) Aluminum gallium nitride [(Al,Ga)N] presents an ideal platform for designing ultraviolet (UV) light emitters across the entire UV spectral range. However, in the deep-UV spectral range (<280 nm) these emitters exhibit very low quantum efficiencies, which in part is linked to the light-polarization characteristics of (Al,Ga)N quantum wells (QWs). In this study, we provide insight into the degree of optical polarization of (Al,Ga)N QW systems operating across the UV-C spectral range by means of an atomistic multiband electronic structure model. Our model not only captures the difference in valence-band ordering in AlN and GaN but it also accounts for alloy-disorder-induced band-mixing effects originating from random alloy fluctuations in (Al,Ga)N QWs. The latter aspect is often not captured in widely employed continuum-based models. The impact of alloy disorder on the electronic structure is studied in terms of Urbach-tail energies, which reflect the broadening of the valence-band density of states due to carrier-localization effects. We find that especially in wider wells, Urbach-tail energies are reduced with increasing carrier densities in the well, highlighting that alloy-disorder-induced carrier-localization effects in (Al,Ga)N QWs are also tightly linked to electrostatic built-in fields. Our calculations show that for QWs designed to emit at the longer-wavelength end of the UV-C spectrum, carrier density and well width are of secondary importance for their light-emission properties, meaning that one observes mainly transverse electrical polarization. However, for (Al,Ga)N QWs with high Al contents, we find that both well width and carrier density will impact the degree of optical polarization. Our calculations suggest that wider wells will increase the degree of optical polarization and may therefore be an option to improve the light-extraction efficiency in deep-UV light emitters. DOI: 10.1103/x5d6-w3cp I. INTRODUCTION The semiconductor alloy aluminum gallium nitride [(Al,Ga)N] has gained significant attention in the past decade as the direct band gap of this semiconductor alloy can in principle span the entire UV range [UV-A (315– 400 nm), UV-B (280–315 nm), and UV-C (200–280 nm)]. Applications that require efficient UV light sources, e.g., light-emitting diodes (LEDs) or lasers, operating over this wide spectral window include water purification, sterilization, plant lighting, sensing, etc. [1]. (Al,Ga)N-based UV light-emitting devices come with significant advantages over widely employed lowand medium-pressure *Contact author: r[email protected] Published by the American Physical Society under the terms of the Creative Commons Attribution 4.0 International license. Further distribution of this work must maintain attribution to the author(s) and the published article’s title, journal citation, and DOI. mercury lamps, as they (i) provide flexible and tunable wavelengths, (ii) do not require toxic mercury, (iii) show no warm-up time, and (iv) exhibit extremely long lifetimes [1,2]. However, when compared to indium gallium nitride [(In,Ga)N], which is used in light emitters operating in the visible spectral range, current UV-light-emitting devices utilizing (Al,Ga)N alloys still exhibit poor external quantum efficiencies, especially in the deep-UV range (<280 nm). Multiple factors contribute to the low efficiency of, e.g., (Al,Ga)N LEDs, which include high threading dislocations and point-defect densities, poor radiative recombination rates, and low light-extraction efficiencies (LEEs) [1]. The latter two aspects in particular are tightly linked to fundamental properties of (Al,Ga)N alloys and heterostructures. First, the thermodynamically stable phase of (Al,Ga)N alloys is the wurtzite crystal structure [1,3] and its lack of inversion symmetry results in a spontaneous polarization field along the wurtzite caxis. In a nitride heterostructure, e.g., an AlxGa1−xN/AlyGa1−yN quantum well (QW), which lies at the heart of UV LEDs, discontinuities 2331-7019/25/24(4)/044084(17) 044084-1 Published by the American Physical Society Exercise ▪Aim: using CSVW & QUDT for semantic annotation of tabular data sets behind the plots and figures in the publication ▪30 data sets containing processed simulation data ▪Largest data sets contained 50+ columns ▪validation using the CLDF/CSVW Python library/CLI ▪Data publication accompanying journal publication on Zenodo DOI zenodo.17226582 DOI 10.1103/x5d6-w3cp Highlight Case study form Computational Materials Science Kickoff Meeting of the GAMM Activity Group on Research Software Engineering and Research Data Management in Mathematics & Mechanics /107 Fig 3. Reprinted under CC BY 4.0 from Phys. Rev. Applied 24, 044084, DOI: https://doi.org/10.1103/x5d6-w3cp 2025-12-05 URBACH-TAIL ENERGIES AND OPTICAL POLARIZATION PHYS. REV. APPLIED 24, 044084 (2025) h-DOS (d-h-DOS) with respect to energy, ε,numerically. The reason for this is that the modification to the DOS due to alloy disorder, the Urbach tail, can be described by an exponential tail, ∝exp (−(ε/Eu));inthe“standard” DOS for an ideal QW (parabolic band), the h-DOS would exhibit a piecewise constant dependence on energy. Taking the derivative of h-DOS with respect to energy ε, one can identify the Urbach-tail region in the d-h-DOS, which is again ∝exp(−(ε/Eu)).Thus,thisallowsusto extract Eufrom the ETBM data. The exponential tail in d-h-DOS is clearly visible in Fig. 2.Theinsetfiguredisplays the natural logarithm of the data, which reflects the expected straight line. From the slope of the line of best fit, Euis extracted. This method is employed below to determine Eufor the considered QW systems at different carrier densities. III. RESULTS In this section, we discuss the Urbach-tail energies and DOPs of (Al,Ga)N QW systems. Special attention is paid to the impact of (i) Al content, (ii) well width, and (iii) carrier density on the results. In Sec. IIIA,westartwiththelower-Al-contentsystem,namely, Al0.48Ga0.52N/Al0.63Ga0.37N QWs, before turning to the higher-Al-content structure, Al0.75Ga0.25N/Al0.90Ga0.10N, in Sec. III B. A. Al0.48Ga0.52N/Al0.63Ga0.37N quantum well systems In Sec. IIIA 1,weanalyzetheUrbach-tailenergiesof Al0.48Ga0.52N/Al0.63Ga0.37NQWsofdifferent widths. In Sec. IIIA 2,theDOPisstudied. 1. Urbach tails Following the procedure outlined in Sec. IIC, in Fig. 3 we display the extracted Urbach-tail energies, Eu, for Al0.48Ga0.52N/Al0.63Ga0.37NQWsofdifferent widths Lwas afunctionofthecarrierdensity,n,inthewell.Inthelowcarrier-density regime, n=1×1018 cm−3,theUrbach-tail energy increases with increasing Lw.Thisfindingindicates that with increasing well width, hole-localization effects increase. It is also revealed in Fig. 3that at least for the Lw=2.3 nm and Lw=3.3 nm systems, the Urbachtail energy Eudecreases with increasing carrier density. Euchanges by approximately 3 meV in both systems when increasing the carrier density from n=1×1018 to n=1×1020, which corresponds to a reduction of approximately 20% in Euand indicates that alloy-disorderinduced carrier localization is amplified by the built-in field. This conclusion is further supported by the data from the Lw=1.3 nm system, in which the potential drop across the QW is smaller when compared to the wider well system. For the narrower well width, Euis approximately constant and may even slightly exceed the FIG. 3. The Urbach-tail energy, Eu,forAl 0.48Ga0.52N/ Al0.63Ga0.37NQWsofwidthLw=1.3 nm (red), Lw=2.3nm (blue), and Lw=3.3 nm (black) as a function of carrier density n. Urbach-tail energies of the systems with Lw=2.3 nm and Lw=3.3 nm at high carrier densities. To gain further insight into the interplay of carrierlocalization effects and the built-in field, we discuss in the following the planar integrated probability density [63], Pm(k)=! ij ! α |cα,m ijk |2,(1) which builds on the ETBM wave function ψm, expressed as ψm=! ijk ! α cα,m ijk φα ijk. (2) The spatial x,y, and zcoordinates of the N=81920 lattice sites in the supercell are labeled above by i,j, and k, respectively. The sp3ETBM basis states are denoted by φα ijk, with α∈"s,px,py,pz#, and cα,m ijk being the expansion coefficients at each of the lattice sites for the single-particle hole state m.Theexpansioncoefficients for a given state m are obtained by diagonalizing the ETBM Hamiltonian. The quantity Pm(k)gives the probability that the hole state ψmis found in the layer specified by kalong the caxis of the system. In Figs. 4(a)–4(f),Pm(k)is displayed for an arbitrarily chosen configuration for the three well widths considered. The data are plotted for low (n=1×1018 cm−3) and high (n=1×1020 cm−3) carrier-density values; the color bar is kept fixed for easier comparison between the different systems and thus to gain a first insight into the question of how screening of the built-in field changes carrier-localization effects in wells with different well widths and carrier densities. The maximum of the color bar is determined by the maximum Pm(k)value found in the Lw=3.3 nm 044084-7 UCUM vocabulary Filename space-delimited columns QUDT Data type Highlight Case study form Computational Materials Science Kickoff Meeting of the GAMM Activity Group on Research Software Engineering and Research Data Management in Mathematics & Mechanics /108 Fig 3. Reprinted under CC BY 4.0 from Phys. Rev. Applied 24, 044084, DOI: https://doi.org/10.1103/x5d6-w3cp 2025-12-05 URBACH-TAIL ENERGIES AND OPTICAL POLARIZATION PHYS. REV. APPLIED 24, 044084 (2025) h-DOS (d-h-DOS) with respect to energy, ε,numerically. The reason for this is that the modification to the DOS due to alloy disorder, the Urbach tail, can be described by an exponential tail, ∝exp (−(ε/Eu));inthe“standard” DOS for an ideal QW (parabolic band), the h-DOS would exhibit a piecewise constant dependence on energy. Taking the derivative of h-DOS with respect to energy ε, one can identify the Urbach-tail region in the d-h-DOS, which is again ∝exp(−(ε/Eu)).Thus,thisallowsusto extract Eufrom the ETBM data. The exponential tail in d-h-DOS is clearly visible in Fig. 2.Theinsetfiguredisplays the natural logarithm of the data, which reflects the expected straight line. From the slope of the line of best fit, Euis extracted. This method is employed below to determine Eufor the considered QW systems at different carrier densities. III. RESULTS In this section, we discuss the Urbach-tail energies and DOPs of (Al,Ga)N QW systems. Special attention is paid to the impact of (i) Al content, (ii) well width, and (iii) carrier density on the results. In Sec. IIIA,westartwiththelower-Al-contentsystem,namely, Al0.48Ga0.52N/Al0.63Ga0.37N QWs, before turning to the higher-Al-content structure, Al0.75Ga0.25N/Al0.90Ga0.10N, in Sec. III B. A. Al0.48Ga0.52N/Al0.63Ga0.37N quantum well systems In Sec. IIIA 1,weanalyzetheUrbach-tailenergiesof Al0.48Ga0.52N/Al0.63Ga0.37NQWsofdifferent widths. In Sec. IIIA 2,theDOPisstudied. 1. Urbach tails Following the procedure outlined in Sec. IIC, in Fig. 3 we display the extracted Urbach-tail energies, Eu, for Al0.48Ga0.52N/Al0.63Ga0.37NQWsofdifferent widths Lwas afunctionofthecarrierdensity,n,inthewell.Inthelowcarrier-density regime, n=1×1018 cm−3,theUrbach-tail energy increases with increasing Lw.Thisfindingindicates that with increasing well width, hole-localization effects increase. It is also revealed in Fig. 3that at least for the Lw=2.3 nm and Lw=3.3 nm systems, the Urbachtail energy Eudecreases with increasing carrier density. Euchanges by approximately 3 meV in both systems when increasing the carrier density from n=1×1018 to n=1×1020, which corresponds to a reduction of approximately 20% in Euand indicates that alloy-disorderinduced carrier localization is amplified by the built-in field. This conclusion is further supported by the data from the Lw=1.3 nm system, in which the potential drop across the QW is smaller when compared to the wider well system. For the narrower well width, Euis approximately constant and may even slightly exceed the FIG. 3. The Urbach-tail energy, Eu,forAl 0.48Ga0.52N/ Al0.63Ga0.37NQWsofwidthLw=1.3 nm (red), Lw=2.3nm (blue), and Lw=3.3 nm (black) as a function of carrier density n. Urbach-tail energies of the systems with Lw=2.3 nm and Lw=3.3 nm at high carrier densities. To gain further insight into the interplay of carrierlocalization effects and the built-in field, we discuss in the following the planar integrated probability density [63], Pm(k)=! ij ! α |cα,m ijk |2,(1) which builds on the ETBM wave function ψm, expressed as ψm=! ijk ! α cα,m ijk φα ijk. (2) The spatial x,y, and zcoordinates of the N=81920 lattice sites in the supercell are labeled above by i,j, and k, respectively. The sp3ETBM basis states are denoted by φα ijk, with α∈"s,px,py,pz#, and cα,m ijk being the expansion coefficients at each of the lattice sites for the single-particle hole state m.Theexpansioncoefficients for a given state m are obtained by diagonalizing the ETBM Hamiltonian. The quantity Pm(k)gives the probability that the hole state ψmis found in the layer specified by kalong the caxis of the system. In Figs. 4(a)–4(f),Pm(k)is displayed for an arbitrarily chosen configuration for the three well widths considered. The data are plotted for low (n=1×1018 cm−3) and high (n=1×1020 cm−3) carrier-density values; the color bar is kept fixed for easier comparison between the different systems and thus to gain a first insight into the question of how screening of the built-in field changes carrier-localization effects in wells with different well widths and carrier densities. The maximum of the color bar is determined by the maximum Pm(k)value found in the Lw=3.3 nm 044084-7 QUDT QUDT QUDT has no meV as unit! https://si-digital-framework.org/SI/units/millielectronvolt Alternative: SI Digital Vocabulary Developer: Bureau International des Poids et Mesures (BIPM) & PTB But much less comprehensive than QUDT Highlight Case study form Computational Materials Science Kickoff Meeting of the GAMM Activity Group on Research Software Engineering and Research Data Management in Mathematics & Mechanics /109 Fig 4. Reprinted under CC BY 4.0 from Phys. Rev. Applied 24, 044084, DOI: https://doi.org/10.1103/x5d6-w3cp 2025-12-05 QUDT QUDT QUDT has no (reduced) probability density as quantity kind! Turn-around: SKOS mapping to IUPAC Goldbook vocabulary But still no reduced probability density = incompatible with unit! No easy support for numbering/indexing of columns = states („State#n“) FINN, O’DONOVAN, KOPRUCKI, and SCHULZ PHYS. REV. APPLIED 24, 044084 (2025) Al0.75Ga0.25N/Al0.90Ga0.10N QW system at a carrier density of n=1×1018 cm−3,whichisdiscussedinSec. IIIB. In general, Figs. 4(a)–4(f) highlight that carrierconfinement effects along the caxis arise from (i) the polarization field and (ii) the well width. However, information about in-plane localization effects is also encoded in Pm(k):forawavefunctionthatisdelocalizedinthe cplane and along the growth direction, Pm(k)will not exhibit strong variations. Therefore, the absence of large variations in Pm(k)for a given state is indicative of the absence of strong carrier-localization effects both due to built-in field and alloy disorder, in the growth plane and along the caxis. For the low carrier density of n=1×1018 cm−3,Figs. 4(a)–4(c) reveal the expected behavior, namely, that the built-in polarization field confines the hole charge density to the lower QW-barrier interface. In Figs. 4(a)–4(c), we also show that the impact of the built-in polarization field is less pronounced in the Lw=1.3 nm QW system [Fig. 4(a)], as the potential drop is much smaller in this system when, e.g., compared to Lw=3.3 nm. In the Lw=2.3 nm [Fig. 4(b)] and Lw=3.3 nm [Fig. 4(c)] QWs, the hole wave functions are clearly confined to much shorter lengths along the caxis than the quantum confinement introduced by the respective QW width (indicated by dashed lines). The confinement lengths here may even be below or close to 1.3 nm. Therefore, at low carrier densities, the hole wave function for the Lw=2.3 nm and Lw=3.3 nm QW experiences strong confining potentials, which are driven by the large potential drop present in these systems. In the Lw=1.3 nm well, the quantum confinement due to the well width plays a larger role. Due to the strong confinement regime in all three cases, local fluctuations in Ga content can then lead to carrier-localization effects, given the lower band gap of GaN, when compared to AlN and the high effective hole masses in general [24]. With increasing carrier density in the well, the builtin field is screened. In Figs. 4(d)–4(f),wedisplayPm(k) at the higher carrier density of n=1×1020 cm−3. The screening effect is of secondary importance for the (a) (b) (c) (d) (e) (f) FIG. 4. The planar integrated probability densities of holes, Pm(k), for an arbitrarily chosen alloy configuration of Al0.48Ga0.52N/Al0.63Ga0.37N QWs of width (a),(d) Lw=1.3 nm, (b),(e) Lw=2.3 nm, and (c),(f) Lw=3.3 nm. The data are displayed at carrier densities of (a)–(c) n=1×1018 cm−3and (d)–(f) n=1×1020 cm−3, for 60 hole states. Index mrefers to the singleparticle hole state number, with m=1 being the ground state, while krefers to the layer in the supercell along the wurtzite caxis. The QW boundaries are indicated by the dashed light blue lines. The color bar is kept the same between the different figures and is determined by the maximum Pm(k)value found in the Lw=3.3 nm Al0.75Ga0.25N/Al0.90Ga0.10N QW system at a carrier density of n=1×1018 cm−3, discussed in Sec. IIIB. 044084-8