WorksMathematics
Learning Morse Lyapunov functions through invertible neural networks
Abstract
Towards data-driven stability certificates of dynamical systems, we are witnessing renewed interest in studying and exploiting the structure of Lyapunov functions. Especially Lyapunov functions that are locally quadratic (i.e., Morse) gained popularity ever since it was understood that they can be written as the norm squared of a diffeomorphism. This topological representation alludes to parametrizations using invertible neural networks. In this note we discuss when and how such functions can be learned. In particular, using neural ODEs (NODEs), we characterize Sobolev approximation of these diffeomorphisms on compact sets and show how this naturally relates to stability. We also connect our results to PLNets and supply numerical experiments for illustration.
Introduction
After pioneering work by Massera [23], Kurzweil [18] and Wilson [33], there is now an extensive theory describing when the existence of a continuous or smooth Lyapunov function is equivalent to the existence of an asymptotically stable attractor under some dynamical system, e.g., see [10, 5, 3, 8]. Towards a data-driven stability analysis, however, one hopes to say more beyond regularity of those Lyapunov functions, e.g., to work with suitable parametric function spaces when searching for a Lyapunov function, see Manchester et al. [22] for a recent overview and Kolter and Manek [17] for work that propelled this field of study. In this article we focus solely on, structure of the Lyapunov function, but note that a lot of work aims to study or identify some unknown dynamical system together with such a stability certificate, e.g., see Harting et al. [11].
In a companion note (Jongeneel [14]), we discuss how the structure of Lyapunov functions helps in uncovering topological features of dynamical systems. We exploit those results below.
Setting
In this note we work with -smooth dynamical systems of the form
for some , and such that some equilibrium point is globally asymptotically stable (GAS). To aid the reader, we briefly recall global asymptotic stability, (e.g., see Bacciotti and Rosier [3], p. 28). Consider (1), and let the corresponding semiflow be denoted by , i.e., for all . We will not be concerned with existence of solutions. A point is an equilibrium point of when or equivalently, is a fixed point of the time- map: for any . Then, the equilibrium point is said to be globally asymptotically stable (GAS) (under ) when: (i) is Lyapunov stable, that is, for any open neighbourhood there is an open set such that ; and (ii) is globally attractive, that is, for all .
It is hard to work directly with the definition of GAS. Luckily, being GAS is equivalent to the existence of a smooth (i.e., -smooth) function that is: (i) coercive (i.e., all sublevel sets are compact); (ii) such that ; and (iii) away from , decreasing strictly under solutions of (1), i.e., for all , e.g., see [3 Thm 2.4]. Functions satisfying properties (i), (ii) and (iii) will be simply called smooth Lyapunov functions (this instead of smooth, strict, proper Lyapunov functions with respect to equilibria). Without loss of generality, in the remainder of this note we focus on . We also remark that we work on for simplicity.
Regarding further notation and structure, following Kvalheim [20], let be the set of -smooth vector fields on , for some , such that is GAS.
Suppose now that we have a family of Lyapunov function candidates, parametrized through via some neural network: . With the aforementioned terminology in place, the central question of this note is to understand precisely when and how these functions can be selected to be Morse. Better yet, we discuss why this is interesting.
We recall that a function is a Morse function when all its critical points (i.e., such that ) are non-degenerate, that is, the Hessian is non-singular. Inspired by a rich set of topological results from Morse theory (e.g., see [24]), one may motivate the study, globally, of Lyapunov functions that are Morse, i.e., for all and some equilibrium . The fact that the Hessian is symmetric positive definite is written as . Such a Lyapunov function will be called a Morse Lyapunov function.
Contribution: In our companion note, inspired by Boumal et al. [6], we studied topological ramifications of working with Lyapunov functions that are Morse [14]. In this work, being inspired by advances in invertible neural networks (e.g., Ishikawa et al. [13]) we detail effectiveness of employing invertible neural networks to learn Morse Lyapunov functions as stability certificates. In particular, we provide (i) an obstruction for the existence of Morse Lyapunov functions (Proposition 1); (ii) as the obstruction is mild, we can provide theory for when NODE-based INNs yield Morse Lyapunov functions on compact sets (Theorem 5, Corollary 7); and (iii) we link these results to PŁ Lyapunov functions, providing context for the distortion parameter (Remark 8).
Simple Lyapunov functions do not suffice, in general
First, when (1) is linear, can be chosen to be quadratic (e.g., for ), globally, e.g., see [30 Thm. 18]. Under hyperbolicity, some arguments survive, locally, for nonlinear systems, e.g., see [21 §9] and [10 §28] for early results or [4] and [11] for recent learning-oriented results.
Globally, one may hope that more can be said beyond -smoothness of , e.g., one may hope that any Lyapunov function can be chosen to be a polynomial. This hope was refuted by Ahmadi et al. [2]. In a similar vein, there are vector fields that render the origin GAS, but provably fail to admit a convex Lyapunov function [15 Ex. III.3]. Moreover, one can find examples where a smooth Lyapunov function exists, but not one that is locally real-analytic, see [3 Prop. 5.2 (b] and [1].
One may continue and hope that Lyapunov functions can always be chosen such that they are Morse, for instance. With the proposition below we illustrate that in general, not much more can be said beyond simple -regularity. The example to have in mind is a planar ODE of the form:
for some . One can use the Lyapunov function to check that is indeed GAS. However, one can show that is not Morse, and in fact, no Morse Lyapunov function exists for (2) ([14 Prop. 3.5]. It seems reasonable that hyperbolicity plays a role here, however, the real problem is having a linearization where the imaginary part cannot be diagonalized ([14 Cor. 3.9]). We generalize this observation below, that is, not only does a locally quadratic Lyapunov function fail for (2), even a Lyapunov function that is sandwiched between positively homogeneous polynomials, in , of equal degree (e.g., quadratic forms), fails to exist cf. the quadratic bounds in Kolter and Manek [17].
Proposition 1 (An obstruction to local homogeneity) Let , for some and . Suppose that has eigenvalues on the imaginary axis that are not semisimple. Suppose that we pick some function such that
for some , and being some neighbourhood of . Then, cannot be a Lyapunov function for under .
Proof. Since has an eigenvalue that is not semisimple, there is some such that . This implies in particular that there is some such that . However, then for a sufficiently small (i.e., to stay within ) one has
where is a remainder term such that . Continuing, after a further decrease in , if necessary, one obtains:
Hence, is increasing along a solution and so it cannot be a valid Lyapunov function.
Note, one may continue this line of arguments and generalize (3) to a subset of functions. We leave this, and in fact a complete characterization, for future work.
A typical non-hyperbolic example that does admit a Morse Lyapunov function is of the form
e.g., see Sastry [28] and Jongeneel [14]. One can check that is a Morse Lyapunov function indeed. In particular, .
Morse Lyapunov functions
Inspired by genericity results (e.g., see Peixoto [26], Hirsch, [12] or Palis and De Melo, [25]), one might expect that Morse Lyapunov functions still provide for a sufficiently rich class of functions. To that end, let be the subset of vector fields such that is a hyperbolic equilibrium point (i.e., for ). In our companion paper we have provided the following converse result, confirming part of what one might expect.
Theorem 2 (On the existence of Morse Lyapunov functions (Jongeneel, [14]). Let , for some and , then, there is a smooth Morse Lyapunov function certifying that is GAS.
A generalization of the Morse Lemma (Milnor, [24]) makes Theorem 2 interesting, that is, through Boumal et al. [6] it follows that Morse Lyapunov functions are elements of
where denotes the group of -smooth diffeomorphisms from to itself (-smooth bijective maps with a -smooth inverse). In fact, one may pick thanks to symmetry in (e.g., see Jongeneel, [14]), that is, the diffeomorphism can be selected to be orientation preserving (like time- maps , for a flow). One may spot that results of this form are closely related to the study of linearizing nonlinear dynamical systems. In general, a diffeomorphism does not suffice, one needs to sacrifice regularity at the equilibrium, see Grüne et al. [9]; Jongeneel [14]; Kvalheim and Sontag [20].
Remark 3 (Regions of attraction that are not ). We have focussed on global asymptotic stability on to keep the exposition simple. However, consider some smooth -dimensional manifold and let be asymptotically stable under some smooth vector field on , with being the corresponding region (basin) of attraction. A topological fact is that is diffeomorphic to (Wilson Jr, 1967, Thm. 2.2). Hence, effectively anything we discuss carries over.
Then, inspired by Wilson Jr (1967), Cheng et al. [7] proposed to use a function , parametrized via BiLipNet (or PLNet for that matter) e.g., see Wang et al. [32], and study Lyapunov functions of the form , which we understand as a predecessor of using (as global smooth PL (Lyapunov) functions, i.e., , for some and all , are a subset of (Boumal et al., [6]). They identified that such a parametrization directly aids in enforcing structure relevant for Lyapunov functions. Since is bi-Lipschitz, the map is invertible, and in contrast to other work, is not enforced to be convex. Convexity may be restrictive (Jongeneel and Schwan, [15]). Then, motivated by finding neural architectures that guarantee for all , Zhong et al. [35] proposed PolarNet in a recent preprint. They effectively recover a subset of the results from Boumal et al. [6], cf. (Zhong et al., [35]). They work with for an appropriate map and claim that if is an affine coupling flow architecture, then any diffeomorphism can be approximated, up to arbitrary precision. The authors appeal to Teshima et al. [31] to substantiate this claim. However, that work provides a universal approximation property with respect to the or sup-norm; and these norms are not of use in the context of Lyapunov inequalities. Then, somewhat paraphrasing, Zhong et al. [35] write that asymptotic stability implies that a Lyapunov function exists that can be accurately modelled using PolarNet. Below we formalize and sharpen this claim.
Topologies to certify stability
Merely universal approximation in some norm does not readily suffice in the context of Lyapunov functions. Suppose that is an unknown Lyapunov function with respect to some dynamical system (1), then any that is close, in the sup-norm, to is by no means a valid Lyapunov function. To get a grip on regularity, define the -norm of a -smooth map , on some (bounded) domain , for simplicity, through
The standard example to have in mind is as follows, given a function , then a perturbation of the form results in . However, . In the context of Lyapunov functions, merely control is futile, e.g., , for a small , does not imply that , despite .
Without going into measure theory, the -norm of such a function is defined through for . The sup-norm, on the other hand, is simply defined as . It is now important to note that both the -norm and the sup-norm do not control the -norm for . Hence, they are of little use in the context of learning Lyapunov functions. This is known, Harting et al. [11] consider -control as well, although not directly by construction, but by appealing to standard universal approximation results.
Universal approximation properties of invertible neural networks
Zhong et al. [35] appeal to Teshima et al. [31] and claim that or sup-norm control suffices to learn Lyapunov functions. We will appeal to more recent results by Ishikawa et al. [13] regarding invertible neural networks (INNs) and control of a Sobolev norm. To follow Ishikawa et al. [13], we need to work with . We return to this in Remark 6.
Let (i.e., some -smooth globally Lipschitz maps) and let be the flow corresponding to for any . Then define the set of diffeomorphisms . Now we may define a set of invertible neural networks (INNs) through
where denotes the affine group, i.e., for some and . Importantly, . Moreover, by composing with some appropriate , one can always force the resulting diffeomorphism to be such that maps to itself.
Then, we say that a set of maps is a -universal approximator for if for any compact set , any and any there is a such that
The next theorem characterizes -universality of neural ODEs (NODEs), that is, using NODEs, one has control over a Sobolev norm.
Theorem 4 ([13 Thm. 44]) Let and suppose that is a -universal approximator for , then is a -universal approximator for .
Concretely, what Theorem 4 means is that if we fix some , then for some appropriate set of globally Lipschitz vector fields , one may hand us any diffeomorphism , a compact set and some , then we know that there is some such that
for and . Appropriate choices of are discussed in [13 p. 19], e.g., MLPs with finite weights and tanh activation functions (since we typically consider ).
As mentioned before, one can always find an -generated diffeomorphism that maps to . Regarding control, let be a compact neighbourhood of and let be such that (5) holds for generating some . Let for , then (e.g., adjust ) is a diffeomorphism and such that
as (5) implies that and as generates some .
Approximation of stability certificates
Combining Theorem 2 and Theorem 4, we can detail approximation quality of NODE-based INNs (i.e., ) to construct Lyapunov functions. Regarding notation, we recall that . Regarding the power of INNs, one should observe that below we effectively only need to be concerned with , other properties are direct, globally.
Theorem 5 (Universal approximation of Morse Lyapunov functions for hyperbolic equilibria) Let , for some and , then
(i) there is a Lyapunov function (i.e., for ) such that for , , certifying that is GAS;
(ii) for any compact neighbourhood of there is an such that if some diffeomorphism satisfies , then is a Lyapunov function satisfying on , thus certifying global asymptotic stability on ; and
(iii) such a can be generated through an invertible neural network , where is a -universal approximator of .
Proof (i) Since , by Theorem 2 there is a Lyapunov function certifying that is GAS, i.e., there is a such that . The fact that can be chosen to satisfy a Lyapunov equation (or inequality for that matter), locally, follows from the proof of [14 Thm. 3.1].
(iii) By Theorem 4, we can find a , generated by some invertible neural network , such that .
(ii) We claim that there is a choice of such that is a Lyapunov function on . As discussed in § 4, if needed, one can redefine the map to be $x \mapsto , i.e., via composition with some . Hence . We remark, with some abuse of notation, that if , then the shifted map satisfies .
Secondly, we see that , for all and in fact, all sublevel sets , for , are compact as those sets are diffeomorphisms of the compact balls .
The remaining property to check is that on . We know that on , plus that . Consider now some compact ball , for some that we select later, and study on the compact set . Let . Then, as . We have that on if on . Now, as and see that
Let , and . If holds for
then . Hence, .
Now we discuss what happens within , better yet, how to select . Observe that , and this satisfies . Following a standard argument (e.g., see the proof of Khalil, [16]Thm. 4.7), it can be shown that there is a and such that on . Indeed, the reason that such a exists follows since , see Jongeneel, [14]Thm. 3.1.
Then, as above, a useful decomposition is as follows:
Akin to , one can set , where one may shrink if needed. In addition, set . Now, if the Sobolev bound holds, one obtains, via the mean value theorem and convexity of (e.g., ), that . Hence, if is such that
then . In its turn, this implies that on .
At last, we conclude by selecting .
Theorem 5 can only handle (as item (iii) relies on Ishikawa et al. [13]), however, we mention the following fact. As such, the case is irrelevant from the learning perspective.
Remark 6 (The case ). Let , for some and suppose that . In this case, the trivial Morse Lyapunov function, i.e., , always works as on . See Sontag and Sussmann [29] for further remarks.
The proof of Theorem 5 reveals that as long as asymptotic stability can be certified using a Morse Lyapunov function, one can still characterize a form of universality. However, on an annulus/shell.
Corollary 7 (Universal approximation of Morse Lyapunov functions). Let , for some and . Suppose, there is a Lyapunov function (i.e., for ), certifying that is GAS. Then,
(i) for any compact neighbourhood of and radius such that , there is an such that if some diffeomorphism satisfies , then is a Lyapunov function satisfying on ; and
(ii) such a can be generated through an invertible neural network , where is a -universal approximator of .
We interpret Corollary 7 as showing that one can learn a Morse Lyapunov function that certifies practical asymptotic stability, cf. [21 §25]. Arguably the most powerful ramification follows when , as in that case, there is always a Lyapunov function . To include , one needs to handle that vanishes at . To that end, Theorem 5 (i) requires more control of , locally. Such a always exists (Jongeneel [14]), but we do not enforce this structure through . Hence, Corollary 7 provides at the moment the most practically relevant guarantees when using vanilla NODEs and . We are not aware of similar results for other contemporary architectures. However, the following remark illustrates that the above extends to PŁ-functions. Exploiting this is future work.
Remark 8 (Equivalence of Morse and PŁ Lyapunov functions on compact sets). It turns out that on compact sets, using smooth PŁ Lyapunov functions is equivalent to using Morse Lyapunov functions. First, -smooth PŁ-functions, with a unique critical point, are Morse by [27 Cor. 2.17] and [6 Thm. 1.1]. The other direction also works, on compact sets. The local argument follows from [27 Cor. 2.11] and continuity allows for an extension to arbitrary compact neighbourhoods. To be somewhat complete, suppose is Morse. Then, around we have for and , but then there is a such that on we have and . Putting it together yields for . Now, if we work on a compact neighbourhood of , we can define and to obtain on . We can conclude by setting .
Numerical experiments
The purpose of this section is to provide an illustration of the above, hence we consider systems that can be visualized (i.e., planar dynamical systems).
We designed the experiments such that all architectures have roughly the same number of parameters. We compare the NODE-INN (e.g., Ishikawa et al. [13] and this work) against Coupling-INN (e.g., Teshima et al. [31]; Zhong et al. [35]), PLNet (via BiLipNet) (e.g., Wang et al. [32]) and a standard MLP (e.g. using a tanh activation). The learning is implemented via Py-Torch, using a scale-free loss function of the form
for some margin , samples and , where the precise form of depends on the architecture. We remark that for any generated through NODE-INN, positive definiteness, properness and the absence of spurious critical points hold by construction. Only the condition needs to generalize beyond the samples. In contrast, for MLP, all defining properties of a Lyapunov function need to generalize.
Experiment A
We start with learning a Lyapunov function for (4). We present some experimental results in Figure 1. One observes similar behaviour when comparing NODE-INN and PLNet, although PLNet is faster, both per iteration and with respect to convergence as a function of total iterations.
Figure 1. Experiment A. NODE-INN generates via with 3 layers and tanh activation functions generating Lipschitz vector fields. Numerical integration is done via RK4. Coupling-INN is effectively the same but uses affine coupling flows instead. The PLNet is implemented with and , whereas MLP is a standard tanh network (2 hidden layers with 20 neurons). We train using 512 samples, and learning rate . The purple levelsets correspond to a Lyapunov function for (4) and the thinner colour-coded curves are some levelsets of found after 500 iterations of Adam. We show the outcomes of 100 independent runs. The thick curves visualize the first run. The training loss displays the mean and the 25th-75th percentile.
Experiment B
In the second experiment we consider a nonlinear system that is smoothly conjugate to a linear hyperbolic system. In particular, let and the matrix be defined through
Now define as the time-1 map of the flow under , for some . Let and see that . Hence, is a twist (as the angular velocity is variable). Now, as is Hurwitz, there is a quadratic form that certifies that is GAS under , but then certifies that is GAS under . This setting fits well to NODE-INN, and under appropriate choices of and , the nonlinearities can be significant. We present some numerical experiments in Figure 2. NODE-INN seems to perform well, whereas other architectures appear to resist convergence. However, we recall that with respect to BiLipNet, one employs -Lipschitz maps , that is, for all . To train the corresponding networks, one sets and , with the distortion (or for ) controlling the conditioning, e.g., the condition number of is bounded from above by . The results in Figure 2 are for some particular , adapting this parameter may improve performance, as shown in Figure 3. This is interesting in the light of Remark 8; to work with PLNet, one needs to find an appropriate , and this number may be large, cf. from Remark 8. At the same time, this restriction of the search space may aid convergence. A precise characterization is left for future work.
Figure 2. Experiment B. NODE-INN, Coupling-INN, PLNet and MLP generate their respective map as in Experiment A. The purple levelsets correspond to a Lyapunov function for the nonlinear system from § 5.2, for and , and the thinner colour-coded curves are some levelsets of found after 1000 iterations of Adam. We show the outcomes of 30 independent runs. Visualization is as in Figure 1.
Figure 3. Experiment B, where the distortion with respect to is adapted.
Conclusion
We have shown that Morse Lyapunov functions are a class of Lyapunov functions with a rich structure that can be exploited both theoretically and algorithmically. Completing the converse theory, also for the other architectures, and providing a rigorous numerical analysis are left for future work.
Generative AI Use
Generative AI tools (Claude Fable 5.2) aided in comprehending some results in the neural networks literature and those tools were used for code and experimental implementation. The author reviewed and verified all AI-assisted content and takes responsibility for the final content of the paper.
References
- [1]Amir Ali Ahmadi and Bachir El Khadir. A globally asymptotically stable polynomial vector field with rational coefficients and no local polynomial Lyapunov function. Syst. Control Lett., 121: 50–53, 2018.arxiv.org/abs/1803.06087
- [2]Amir Ali Ahmadi, Miroslav Krstic, and Pablo A Parrilo. A globally asymptotically stable polynomial vector field with no polynomial Lyapunov function. In Proc. IEEE CDC, pages 7579–7580, 2011.DOI
- [3]Andrea Bacciotti and Lionel Rosier. Lyapunov Functions and Stability in Control Theory. Springer Science & Business Media, Berlin, 2005.
- [4]Matthieu Barreau and Nicola Bastianello. Learning maximal regions of attraction on compact sets using Taylor Lyapunov networks. arXiv preprint arXiv:2408.17246v3, 2024.arxiv.org/abs/2408.17246
- [5]Nam Parshad Bhatia and Giorgio P Szegö. Stability Theory of Dynamical Systems. Springer, Berlin, 1970.DOI
- [6]Nicolas Boumal, Christopher Criscitiello, and Quentin Rebjock. Smooth, globally Polyak–Łojasiewicz functions are nonlinear least-squares. arXiv preprint arXiv:2604.07972, 2026.arxiv.org/abs/2604.07972
- [7]Jing Cheng, Ruigang Wang, and Ian R Manchester. Learning stable and passive neural differential equations. In Proc. IEEE CDC, pages 7859–7864, 2024.arxiv.org/abs/2404.12554
- [8]Albert Fathi and Pierre Pageault. Smoothing Lyapunov functions. T. Am. Math. Soc., 371(3):1677–1700, 2019.
- [9]Lars Grüne, Eduardo D Sontag, and Fabian R Wirth. Asymptotic stability equals exponential stability, and ISS equals finite energy gain—if you twist your eyes. Syst. Control Lett., 38(2):127–134, 1999.
- [10]Wolfgang Hahn. Stability of Motion. Springer, Berlin, 1967.DOI
- [11]Alice Harting, Karl Henrik Johansson, Sophie Tarbouriech, and Matthieu Barreau. Locally stable neural ODEs with characterized region of attraction. arXiv preprint arXiv:2606.19109, 2026.
- [12]Morris W Hirsch. Differential Topology. Springer, New York, 1976.DOI
- [13]Isao Ishikawa, Takeshi Teshima, Koichi Tojo, Kenta Oono, Masahiro Ikeda, and Masashi Sugiyama. Universal approximation property of invertible neural networks. J. Mach. Learn. Res., 24(287): 1–68, 2023.arxiv.org/abs/2204.07415
- [14]Wouter Jongeneel. On global Lyapunov functions being Morse. arXiv preprint arXiv:2610.05626, 2026b.arxiv.org/abs/2610.05626
- [15]Wouter Jongeneel and Roland Schwan. On continuation and convex Lyapunov functions. IEEE Trans. Automat. Contr., 69(10):6895–6906, 2024.DOI
- [16]Hassan K Khalil. Nonlinear Systems. Prentice Hall, Upper Saddle River, 2002.
- [17]J Zico Kolter and Gaurav Manek. Learning stable deep dynamics models. In Proc. NeurIPS, volume 32, pages 11128–11136, 2019.arxiv.org/abs/2001.06116
- [18]Jaroslav Kurzweil. On the inversion of Ljapunov’s second theorem on stability of motion. AMS Transl. Ser. 2, 24:19–77, 1963.
- [20]Matthew D Kvalheim and Eduardo D Sontag. Global linearization of asymptotically stable systems without hyperbolicity. Syst. Control Lett., 203:106163, 2025.arxiv.org/abs/2502.07708
- [21]Joseph La Salle and Solomon Lefschetz. Stability by Liapunov’s Direct Method with Applications. Academic Press, New York, 1961.
- [22]Ian R Manchester, Ruigang Wang, and Nicholas H Barbara. Neural networks in the loop: Learning with stability and robustness guarantees. Annu. Rev. Control Robot. Auton. Syst., 9, 2026.DOI
- [23]Jose L Massera. Contributions to stability theory. Ann. Math., 64(1):182–206, 1956.DOI
- [24]John Milnor. Morse Theory. Princeton University Press, Princeton, 1963.
- [25]Jacob Jr Palis and Wellington De Melo. Geometric Theory of Dynamical systems: an Introduction. Springer, New York, 1982.
- [26]Mauricio M Peixoto. On an approximation theorem of Kupka and Smale. J. Differ. Equ., 3(2):214–227, 1967.DOI
- [27]Quentin Rebjock and Nicolas Boumal. Fast convergence to non-isolated minima: four equivalent conditions for C^2 functions. Mathematical Programming, 213(1):151–199, 2025.
- [28]Shankar Sastry. Nonlinear Systems. Springer, New York, 1999.
- [29]Eduardo Sontag and Hector Sussmann. Remarks on continuous feedback. In Proc. IEEE CDC, pages 916–921, 1980.DOI
- [30]Eduardo D Sontag. Mathematical Control Theory: Deterministic Finite Dimensional Systems. Springer, New York, 1998.
- [31]Takeshi Teshima, Isao Ishikawa, Koichi Tojo, Kenta Oono, Masahiro Ikeda, and Masashi Sugiyama. Coupling-based invertible neural networks are universal diffeomorphism approximators. In Proc. NeurIPS, pages 3362–3373, 2020.arxiv.org/abs/2006.11469
- [32]Ruigang Wang, Krishnamurthy Dj Dvijotham, and Ian Manchester. Monotone, bi-Lipschitz, and Polyak-Lojasiewicz networks. In Proc. ICML, pages 50379–50399. PMLR, 2024.arxiv.org/abs/2402.01344
- [33]F Wesley Wilson. Smoothing derivatives of functions and applications. T. Am. Math. Soc., 139: 413–428, 1969.DOI
- [35]Yuan Zhong, Jiaxin Cheng, Hefu Ye, and Yicong Zhou. Polarnet: Single-minima neural network for modeling Lyapunov functions. arXiv preprint arXiv:2605.09546, 2026.arxiv.org/abs/2605.09546


