PaperlayerView exact revision on arXiv

Source-linked AI summary

On the differential of the exponential map

Debanjan Mallik

arXiv:2609.04762v1eess.SYmath.OC

TL;DR

The paper addresses how to compute the time derivative of the matrix exponential and how to formulate the differential of the exponential map. It derives equivalent summation, integral, Lie-bracket, and adjoint representations, extends the result to parameterized operator families, and proves existence of the Fréchet derivative. The concluding result is an explicit and unified framework for these derivative formulas.

  • Problem

    The paper studies the time derivative of the matrix exponential and its relation to the differential of the exponential map.

  • Method

    The paper uses Taylor expansions, Euler beta and gamma integrals, Lie brackets, adjoint operators, an extension to parameterized operator families, and an alternative proof.

  • Results

    The paper derives equivalent integral, Lie-bracket, and adjoint forms, verifies the family extension, and establishes existence of the Fréchet derivative.

  • Takeaways & Limitations

    The results provide explicit formulas for the Fréchet differential of the exponential map and a deductive beta-integral approach without assuming the answer’s functional form.

Abstract

from arXiv · show

We study the time derivative of the matrix exponential $B(t)=\mathrm{exp}(A(t))$, where $A(t)$ is a time-parametrized curve in $\mathrm{Mat}(n)$. Starting from the Taylor series expansion, we derive a nested summation formula, which is then reformulated into a double summation. This expression is converted into an integral representation using Euler's beta function, and further expressed in terms of Lie brackets and the adjoint representation. Later, we extend this to a family of problems and verify a well-known result. Finally, we establish connections to the Gateaux and Fréchet derivatives, proving the latter's existence. These results offer a unified and explicit framework for understanding the derivative of the exponential map.

I. INTRODUCTION

The report studies the time derivative of the matrix exponential on Mat(n), situating it within the Lie-theoretic exponential map. It develops this problem through summation, integral, Lie-algebraic, and derivative-based formulations.

  • For matrix Lie groups, the Lie-theoretic exponential map is represented by the matrix exponential.
  • The central problem is computing the time derivative of the matrix exponential for a time-parametrized matrix curve.
  • The report first derives nested and double summations, then converts them into an integral representation using Euler’s beta function.
  • It further expresses the result using Lie brackets and the adjoint representation, extends the problem to a family of problems, and connects it to Gateaux and Fréchet derivatives.

II. NESTED AND DOUBLE SUMMATION FORMS

The derivative of exp(A(t)) is obtained by differentiating the Taylor series term by term and applying the product rule to noncommuting matrix factors. Reindexing the resulting nested sums produces a double-summation form.

  • Absolute and uniform convergence of the matrix exponential and its partial derivatives permits differentiation inside the Taylor series.
  • Because matrices generally do not commute, the chain rule is replaced by the product rule applied to the k factors of A(t)^k.
  • Differentiating each factor in turn yields terms with dot A(t) inserted at different positions among the remaining factors.
  • Reindexing with j=i−1 and then introducing independent indices p and q gives k=p+q+1.
  • The nested summation is thereby rewritten as a double summation involving A(t)^p dot A(t) A(t)^q divided by (p+q+1)! .
  • The factorial structure of the double sum motivates using the beta-gamma relationship to replace the summation with exponential terms.

III. BETA AND GAMMA FUNCTIONS: EULER INTEGRALS OF FIRST AND SECOND KIND

The factorial coefficient in the double summation is identified with an Euler beta integral through the gamma function. This converts the series into the final integral form.

  • The gamma function is introduced as Euler’s integral of the second kind, while the beta function is introduced as Euler’s integral of the first kind.
  • The beta-gamma relationship connects the integral with gamma-function values at positive integers.
  • Choosing x=q+1 and y=p+1 identifies the beta integral with the factorial coefficient involving (p+q+1)! .
  • After this identification, the double sum is replaced by power-series expansions for exponentials.
  • The resulting expression is presented as the final integral form for the derivative.

IV. LIE BRACKETS/ADJOINT ACTIONS FORM

The integral representation is recast in terms of Lie brackets and little-ad operators. The resulting adjoint-action formula is identified as the intended form of the derivative.

  • The section derives the time derivative of the matrix exponential in terms of adjoint actions and defines the little-ad operation.
  • The adjoint action of a Lie algebra on itself can be expressed for matrices using the Lie bracket.
  • Powers of little ad correspond to nested commutators, such as ad_X(ad_XY)=[X,[X,Y]].
  • The integral representation is transformed through a change of variables and subsequent algebraic identities into an adjoint-operator expression.
  • Uniform convergence justifies interchanging summation and integration when deriving the operator series.
  • The resulting little-ad formula is presented as the intended form and is associated with earlier proofs by Schur and Poincaré.

A. An Alternative Proof

An alternative proof verifies the matrix-exponential derivative under continuous differentiability by introducing an auxiliary construction, differentiating it, and using cancellation and the fundamental theorem of calculus. The resulting expression exactly matches the earlier formula.

  • A. An Alternative Proof: Assuming A(t) is continuously differentiable, the proof introduces an auxiliary construction to establish the derivative identity.The argument proceeds by differentiating with respect to an auxiliary parameter and comparing both sides.
  • A. An Alternative Proof: Mixed partial derivatives commute, causing the first and third right-hand-side terms to cancel.This reduces the verification to the remaining terms.
  • A. An Alternative Proof: The fundamental theorem of calculus completes the derivation of the target expression.The proof uses the integral relation to recover the desired identity.
  • A. An Alternative Proof: The alternative proof obtains exactly the result derived earlier.Thus, the independently constructed verification agrees with the previous formula.

V. EXTENSION TO A FAMILY OF PROBLEMS

The integral formula is extended from the time-dependent matrix problem to a parameter-dependent family involving an operator H(λ). The extension is verified by showing that both sides satisfy the same relation and initial data.

  • V. EXTENSION TO A FAMILY OF PROBLEMS: The section extends the integral form to a family of problems where the operator H depends on a parameter λ.This is presented as a lemma attributed to Wilcox.
  • V. EXTENSION TO A FAMILY OF PROBLEMS: The factor −β makes the extended equation more general than the original integral form, with time t corresponding to λ.A(t) plays the role of the parameter-dependent matrix function H(λ).
  • V. EXTENSION TO A FAMILY OF PROBLEMS: The verification proceeds by accounting for the negative sign and differentiating the relevant relation with respect to β.The proof explicitly compares the left- and right-hand sides.
  • V. EXTENSION TO A FAMILY OF PROBLEMS: Zero initial data imply that the difference F−G is identically zero, establishing equality of the two sides.Uniqueness of the solution with zero initial data completes the verification.

VI. CONNECTION TO GATEAUX AND FR´ECHET DERIVATIVES

The paper connects the matrix-exponential formula to Gateaux and Fréchet differentiation. It identifies the integral expression as the Gateaux differential and proves that the remainder is o(||H||), establishing existence of the Fréchet derivative.

  • VI. CONNECTION TO GATEAUX AND FR´ECHET DERIVATIVES: The Gateaux differential is defined through directional limits, while the Fréchet differential requires a linear bounded approximation with a smaller-order remainder.The paper also states that an existing Fréchet differential equals the corresponding Gateaux differential.
  • VI. CONNECTION TO GATEAUX AND FR´ECHET DERIVATIVES: For the exponential map on Mat(n), the paper evaluates the Gateaux differential at A in the direction H.The increment is introduced through A(ε)=A+εH.
  • VI. CONNECTION TO GATEAUX AND FR´ECHET DERIVATIVES: The Fréchet derivative exists after the error term is shown to be o(||H||) as ||H|| tends to zero.The paper therefore promotes the Gateaux expression to the Fréchet derivative.
  • VI. CONNECTION TO GATEAUX AND FR´ECHET DERIVATIVES: The matrix expansion is reorganized by indexing the terms over j and ℓ, with the number of terms determined from their allowed ranges.The argument uses a submultiplicative matrix norm to control the remainder.
  • VI. CONNECTION TO GATEAUX AND FR´ECHET DERIVATIVES: The first term in the expansion of (A+H)^k−A^k is identified with the beta-integral form, while the remaining terms define R(H).This separates the candidate differential from its remainder.

VII. CONCLUDING REMARKS

The report derives the time derivative of exp(A(t)) using Taylor expansions, Euler integrals, and Lie algebraic tools, and establishes equivalent integral, Lie-bracket, and adjoint-operator forms. It also gives explicit formulas for the Fréchet differential.

  • VII. CONCLUDING REMARKS: The time derivative of the matrix exponential is derived using Taylor expansions, Euler integrals, and Lie algebraic tools.These ingredients form the report's main derivational framework.
  • VII. CONCLUDING REMARKS: The integral, Lie-bracket, and adjoint-operator formulations are shown to be equivalent.The conclusion presents these as alternative expressions of the same derivative result.
  • VII. CONCLUDING REMARKS: The beta-integral approach is deductive and does not assume the functional form of the answer.The conclusion contrasts this with the differential-equation approach attributed to Rossmann and Wilcox.
  • VII. CONCLUDING REMARKS: The results provide explicit formulas for the Fréchet differential of the exponential map.This extends the derivative analysis beyond the time-parametrized formulation.
Loading 2609.04762v1…