Tensor Forms of Derivatives of Matrices and their applications in the Solutions to Differential Equations
Yiran Xu, Guangbin Wang, Changqing Xu · 2025
Read Tensor Forms of Derivatives of Matrices and their applications in the Solutions to Differential Equations by Yiran Xu, Guangbin Wang, Changqing Xu (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-08429.txt.
Introduction
Yiran Xu and Changqing Xu build a unified calculus for matrix-valued functions in a 2025 maths paper. Such functions run through control theory, imaging, and model reduction. Their rates of change are often handled case by case. The pair set out to make the rules systematic, once and for all.
The derivative of one matrix with respect to another is a fourth-order tensor. With that in hand, whole systems of equations can be written in a compact tensor form. The solution mirrors the scalar case: an exponential of the system tensor, times a constant. A partial Tucker split then shrinks big tensor equations for fast use. Engineers with huge coupled systems need cuts that keep the structure. A tensor-native calculus makes those cuts routine.
Catalog record
| CREATOR | Yiran Xu, Guangbin Wang, Changqing Xu |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 81 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-08429
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Tensor Forms of Derivatives of Matrices and their applications in the Solutions to Differential Equations Yiran Xu, Guangbin Wang, Changqing Xu Abstract We introduce and extend the outer product and contractive product of tensors and matrices, and present some identities in terms of these products. We offer tensor expressions of derivatives of tensors, focus on the tensor forms of derivatives of a matrix w.r.t. another matrix. This tensor form makes possible for us to unify ordinary differential equations (ODEs) with partial differential equations (PDEs), and facilitates solution to them in some cases. For our purpose, we also extend the outer product and contractive product of tensors (matrices) to a more general case through any partition of the modes, present some identities in terms of these products, initialize the definition of partial Tucker decompositions (TuckD) of a tensor, and use the partial TuckD to simplify the PDEs. We also present a tensor form for the Lyapunov function. Our results in the products of tensors and matrices help us to establish some important equalities on the derivatives of matrices and tensors. An algorithm based on the partial Tucker decompositions (TuckD) to solve the PDEs is given, and a numerical example is presented to illustrate the efficiency of the algorithm. keywords: Contractive product; derivative; linear ordinary differential equation; outer product; Lyapunov function. AMS Subject Classification : 53A45, 15A69. 1 Introduction The theory of differential equations (DEs) was originated earlier in the 17th century for the need to model dynamic systems in astronomy, physics and geometry. The connection between the theory of DEs and linear and multilinear algebra is deep and multifaceted. The linear and multilinear algebra are foundational to the formulation, solution, and interpretation of DEs. While DEs are primarily studied using functional analysis, linear and multilinear algebra do provide essential tools for solving, and analyzing these equations. A system of linear ordinary differential equations (ODEs) can be written as \frac{d\textbf{x}}{dt}=A\textbf{x}, where \textbf{x}\in\mathit{R}^{n} is a vector and A\in\mathit{R}^{n\times n} is a constant matrix. The solution to this equation involves some central concepts in linear algebra such as the eigenvalues, eigenvectors, and matrix exponentials e^{At} . Now we may ask: what if x or t is (or both are) replaced by a matrix or a higher order tensor ? In this paper, we will define the derivative of a tensor (matrix) with respect to another tensor. We focus on the tensor forms of derivatives of a matrix X w.r.t. another matrix T . This tensor form makes possible for us to unify the ODEs and partial differential equations (PDEs), and facilitates solution to them in some cases. For our purpose, we first extend the outer product and contractive product of tensors and matrices to more general case through any partition of the modes, present some identities in terms of these products, we define the partial Tucker decompositions (TuckD) of a tensor, and use the partial TuckD to simplify the PDEs. We also present a tensor form for the Lyapunov function. Our results in the products of tensors and matrices help us to establish some important equalities on the derivatives of matrices and tensors. Tensors, as the central concept in multilinear algebra, are frequently used to describe PDEs in curvilinear coordinates. Tensor analysis is used to handle nonlinear terms in PDEs (e.g., u\nabla u , in fluid dynamics), which are inherently multilinear. Solutions to PDEs (e.g., heat/wave equations) are expressed as series of eigenfunctions of differential operators (e.g., \Delta u=\lambda u ), analogous to diagonalizing matrices in linear algebra. While tensors are not central to the theory of differential equations, they become powerful for ODEs on manifolds. For partial differential equations (PDEs), tensors are indispensable tools for formulating covariant, geometrically meaningful, and computationally structured equations. To describe the motion constrained to some manifolds (e.g., rigid body rotation, robotic arms), we need some tools e.g. Riemannian metric tensors (matrices) to define distances or inner products which is crucial for kinetic energy, and the curvature tensor (third order tensor) governs the manifold’s geometry which the ODE solution respects. Also in the advanced geometric theory of ODEs, higher order derivatives, often described in tensor form, are treated as coordinates on a manifold, and the geometric structures on the jet spaces are also expressed by tensors. The concept of tensor can be dated back to the 19th century when Cauchy (1822) developed the stress tensor (second order tensor, i.e., a matrix), to describe internal forces in materials. Small order tensors (vectors and matrices) were formalized by Cauchy, Grassmann, and Hamilton in the mid-19th century. Grassmann (1844) introduced the idea of multilinear algebra which was the first explicit use of tensor-like objects in physics. Tensors are fundamental and ubiquitous in data sciences, mechanics, physics and chemistry, they are also used in PDEs especially those arising in continuum physics and geometry. For more detail on the development of tensor theory, we refer to [ 15 ] . An essential power of tensors lies in its invariance of the algebraic form under coordinate transformations and the direct representation of physical quantities and geometric structures independent of coordinate systems. For example, the Cauchy stress tensor (matrix) relates surface forces to directions, with governing equations inherently containing tensor derivatives, and strain tensors can be used to measure deformations (symmetric matrices). In electromagnetism (Maxwell’s equations), the E-M Field tensor, as a second order antisymmetric tensor (matrix), unifies electric (E) and magnetic (B) fields. In Einstein’s general relativity, the equation G_{uv}=(8\pi G/c^{4})T_{uv} relates the curvature of spacetime (described by Einstein tensor G_{uv} derived from the 4-order Riemann curvature tensor) to the distribution of matter/energy. Tensors can be used to model order reduction of high-dimensional PDEs arising in quantum chemistry, stochastic PDEs, etc., to mitigate the ”curse of dimensionality”. To make some algorithms more efficient, we usually use tensor algebra libraries to exploit inherent structure (symmetry, sparsity) for efficient storage and computation. The term tensor was put in use early in 1837 by physicists and has been popular since 1925 when Albert Einstein used it to describe general relativity. It can be found in chemometric, data science, image analysis, medical science, psychology and quantum physics, etc.. A tensor can be regarded as an extension of matrices. The main topics in tensor analysis include tensor decompositions, spectral theory, nonnegative tensors,symmetric tensors [ 5 , 13 , 11 ] and structured tensors [ 13 , 11 , 18 ] . Even though tensors can be found in many areas, its appearance in ODE systems is rare if not missing. Tensors are coordinate-invariant, making them ideal for modeling phenomena in physics (relativity, continuum mechanics), engineering (stress analysis, fluid dynamics), machine learning (neural networks, data representation), computer graphics (lighting, deformations), quantum mechanics (spinors, entanglement). Tensors obey strict rules when coordinates change (covariant/contravariant behavior). They describe relationships across multiple dimensions simultaneously. In general relativity, the metric tensor and stress-energy tensor are used to describe spacetime curvature and encodes mass-energy distribution respectively. In continuum mechanics, the Cauchy stress tensor and the strain tensors are utilized to model forces in materials and describe material deformation respectively. In electromagnetism (Maxwell’s equations), the electromagnetic field tensor can be used to unifies E and B fields. In machine learning and data science, weights in a neural network