Related Experiment Video
Updated: Jul 17, 2025

Optimization of Synthetic Proteins: Identification of Interpositional Dependencies Indicating Structurally and/or Functionally Linked Residues
Published on: July 14, 2015
Local invertibility and sensitivity of atomic structure-feature mappings.
Sergey N Pozdnyakov1, Liwei Zhang2, Christoph Ortner2
1Laboratory of Computational Science and Modelling, Institute of Materials, Federal Institute of Technology (EPFL) CH-1015 Lausanne, Lausanne, 1015, Switzerland.
Machine learning for atomic simulations requires understanding how atomic coordinates map to descriptors. This study reveals desirable symmetry-induced singularities and problematic spurious singularities in these mappings, offering tools to identify them.
Area of Science:
- Computational chemistry
- Materials science
- Machine learning
Background:
- Machine learning is increasingly used in atomic-scale simulations.
- Understanding the mathematical mapping from atomic coordinates to descriptors is crucial.
- Symmetric descriptors are essential for enforcing physical constraints in models.
Purpose of the Study:
- To analyze the sensitivity of the mapping between atomic coordinates and descriptors.
- To quantify this sensitivity using singular value decomposition (SVD) of the Jacobian.
- To identify sources of singularities in descriptor representations.
Main Methods:
- Singular value decomposition (SVD) of the Jacobian matrix.
- Analysis of mapping sensitivity to atomic displacements.
- Investigation across different atomic configurations and descriptor choices.
Main Results:
- Symmetry and smoothness create desirable null singular values in the mapping Jacobian.
- Spurious singularities arise from descriptor incompleteness, where distinct structures map to the same feature vector.
- Aggressive truncation of basis sets can introduce additional spurious singularities.
Conclusions:
- Constructing symmetric atomic representations involves subtle mathematical challenges.
- Singularities can be both physically desirable (enforcing symmetry) and problematic (due to incompleteness or truncation).
- Provides conceptual and numerical tools for identifying and analyzing singularities in atomic representations.
Related Concept Videos
Woodward–Hoffmann Selection Rules and Microscopic Reversibility
¹H NMR: Interpreting Distorted and Overlapping Signals
As Δν decreases and the signals move closer, the doublets appear increasingly distorted. The intensities of the inner lines increase at the cost of those of the outer lines as the signals are...
Conserved Binding Sites
Binding sites are often located in large pockets, and if their location on a protein’s surface is unknown, it can be predicted using various approaches. The energetic method computationally...
¹H NMR Chemical Shift Equivalence: Homotopic and Heterotopic Protons
Resonance and Hybrid Structures
Resonance Structures and Resonance Hybrids
The Lewis structure of a nitrite anion (NO2−) may actually be drawn in two different ways, distinguished by the locations of the N–O and N=O bonds.
Inductive Effects on Chemical Shift: Overview

