--- title: Hyers-Ulam Stability Fundamentals url: https://www.emergentmind.com/topics/hyers-ulam-stability type: topic --- # Hyers-Ulam Stability Fundamentals Hyers-Ulam stability is the property that an approximate solution of an equation can be approximated by an exact solution, with the approximation error controlled linearly by the size of the defect. In the classical tradition, Ulam asked whether approximate homomorphisms are near exact homomorphisms, and Hyers proved this for additive maps between Banach spaces; subsequent work has extended the same paradigm to functional equations, differential and difference equations, dynamic equations on time scales, operator equations, random dynamics, and more recent settings such as locally convex cones and weighted graphs [1506.06521][2602.03880]. ## 1. Definition and principal variants In its standard form, Hyers-Ulam stability asserts that if a function satisfies a governing equation up to a uniformly bounded error \(\varepsilon\), then there exists an exact solution within \(K\varepsilon\), for some constant \(K>0\). For instance, for the second-order time-scale equation \[ x^{\Delta\Delta}(t)+\alpha x^\Delta(t)+\beta x(t)=0, \] the stability statement is: whenever \(y\in C_{rd}^{\Delta^2}[a,b]_{\mathbb T}\) satisfies \[ \bigl|y^{\Delta\Delta}+\alpha y^\Delta+\beta y\bigr|\le \varepsilon, \] there exists an exact solution \(u\) such that \[ |y-u|\le K\varepsilon \] on the interval [1008.3726]. For linear operators, the same idea admits an equivalent kernel-distance formulation. If \(T\) is linear, Hyers-Ulam stability is equivalent to the existence of \(K>0\) such that for every \(x\in D(T)\) there exists \(x_0\in N(T)\) with \[ \|x-x_0\|\le K\|Tx\|. \] In Hilbert-space operator theory, the infimum of such constants is denoted \(K_T\) [1211.0192]. The modern literature also contains systematic variants. A recent generalization replaces the usual single norm by a mixed norm pair: in \((L^p,L^q)\) Hyers-Ulam stability, the residual of a pseudosolution is measured in \(L^q\), while the distance to an exact solution is measured in \(L^p\) [2409.14108]. Another line of work explicitly studies Hyers-Ulam-Rassias stability and \(\sigma\)-semi-Hyers-Ulam stability; for Bessel and modified Bessel equations with initial conditions, sufficient conditions for these forms of stability, as well as for ordinary Hyers-Ulam stability, were obtained by integral techniques and majorations [1805.06704]. Two points recur throughout the subject. First, the relevant norm or topology is part of the definition rather than a superficial choice. Second, existence of a nearby exact solution does not automatically imply uniqueness; several papers isolate “stability with uniqueness” as a strictly stronger property [2401.04516]. ## 2. Structural proof mechanisms Although the subject originated in functional equations, the technical machinery now depends strongly on the ambient category. One major method is averaging by invariant means. For generalized additive equations on groups and homogeneous spaces, the correction from an approximate pair \((f,h)\) to an exact pair \((F,H)\) is obtained by averaging defect functions over the acting group, using amenability or strong amenability; in that setting the exact equations \[ H(yz)=H(y)+H(z), \qquad F(xy)=F(x)+H(y) \] are recovered together with the bounds \[ \|H-h\|\le \delta, \qquad \|F-f\|\le 2\delta \] [1506.06521]. A second major method is fixed-point theory. For nonlinear Volterra integral equations, \[ y(x)=\int_a^x V(x,\tau,y(\tau))\,d\tau, \] a generalized Diaz-Margolis theorem on a complete generalized metric space yields existence, uniqueness, and Hyers-Ulam stability under a variable contractive condition expressed by an \(MT\)-function. The resulting estimate is \[ |y(x)-y_0(x)|\le \frac{\theta}{1-\delta} \] for all \(x\in[a,b]\) [1503.07967]. A third and especially pervasive method is factorization into first-order problems. In second-order dynamic equations on time scales, the operator is factorized through characteristic roots or Riccati reduction; in higher-order Cauchy-Euler dynamic equations, the factorization \[ \sum_{k=0}^{n} a_k M_k y(t)= \prod_{k=1}^{n}(\varphi D-\lambda_k I)y(t) \] reduces the problem to a chain of first-order Hyers-Ulam estimates [1008.3726][1212.4163]. The same pattern reappears in discrete Hill-type equations, where second- and third-order \(h\)-difference operators are controlled by composing first-order periodic-coefficient stability results [2303.10072]. In dynamical settings, the dominant mechanism is hyperbolicity. For random linear cocycles with tempered exponential dichotomy, adapted norms and a Green operator on sequence spaces yield a random shadowing theorem; under uniform exponential dichotomy, this becomes a genuine Hyers-Ulam stability result for the linear random dynamics [1909.08707]. This suggests that, in many modern formulations, Hyers-Ulam stability is best understood as an admissibility or shadowing property induced by an underlying splitting or inverse estimate. ## 3. Differential, difference, and dynamic equations Large parts of the literature concern linear equations whose stability can be characterized in terms of spectral, asymptotic, or Floquet-type data. For second-order linear dynamic equations on time scales, Hyers-Ulam stability holds for \[ x^{\Delta\Delta}(t)+\alpha x^\Delta(t)+\beta x(t)=0 \] when the characteristic equation \(\lambda^2+\alpha\lambda+\beta=0\) has two distinct positive roots; the inhomogeneous constant-coefficient equation and a variable-coefficient equation with Riccati reduction are treated in the same framework [1008.3726]. Higher-order Cauchy-Euler dynamic equations on time scales admit Hyers-Ulam stability under the nonvanishing conditions \[ |\varphi(t)|\ge A>0, \qquad \varphi(t)+\lambda_k\mu(t)\neq 0, \] and the final bound is obtained by iterating first-order estimates [1212.4163]. For singular differential equations, the first-order equation \[ t^\gamma y'(t)+z\,y(t)=0 \] on \((0,\infty)\) has Hyers-Ulam stability if and only if \(\Re z\neq 0\); factorizable higher-order equations \[ \sum_{k=0}^n \alpha_{n-k}(t^\gamma D)^k y(t)=0 \] are stable on \((0,\infty)\) if and only if \(\Re z_k\neq 0\) for each first-order factor \((t^\gamma D+z_k I)\) [1307.3287]. For constant-coefficient \(2\times 2\) systems \[ x'(t)=Ax(t), \qquad t\in\mathbb R, \] the criterion is sharp: the system is Hyers-Ulam stable on \(\mathbb R\) if and only if both eigenvalues of \(A\) have nonzero real part. In several cases the best Hyers-Ulam constant is \[ K_{\min}=\|A^{-1}\|_\infty \] [2203.13065]. A related nonautonomous finite-dimensional theory shows that discrete two-sided dynamics \(x_{n+1}=A_nx_n\) is Hyers-Ulam stable with uniqueness if and only if it admits an exponential dichotomy on \(\mathbb Z\), while continuous-time nonautonomous equations are characterized by summable dichotomy or summable trichotomy rather than exponential boundedness assumptions [2401.04516]. Discrete and quantum equations furnish sharper parameter-dependent phenomena. For the first-order Cayley quantum equation \[ D_qx(t)-\omega(\eta x(qt)+(1-\eta)x(t))=0, \] Hyers-Ulam stability holds when \(q>1\), \(\omega\neq 0\), and \(0\le \eta<\tfrac12\), and the best constant is \[ K=\frac1{|\omega|}, \] independent of both \(q\) and \(\eta\); if \(\eta=\tfrac12\), Hyers-Ulam stability fails for every complex coefficient [2005.05122]. For the alternating-step time scale \(\mathbb T_{\alpha,\beta}\), the first-order equation \(x^\Delta=\lambda x\) admits a detailed case-by-case classification, with minimal constant known in the positively regressive regime \[ K_{\min}=\frac1{|\lambda|} \] but generally unresolved in sign-changing regimes [1807.03859]. For the discrete Hill equation \[ \Delta_h^2 y(t)+[\Delta_h\lambda(t)-\lambda(t)\lambda(t+h)]y(t)=0, \] periodic coefficients lead to an explicit constant \[ K=K_0(\lambda)K_0(-\lambda) \] under the nonresonance conditions \(0<|e_{\pm\lambda}(nh)|\neq 1\) [2303.10072]. These results also delimit the theory’s scope. Stability may depend on the domain, as in singular equations on \((0,1)\) versus \((1,\infty)\) [1307.3287], or on excluding dynamically dangerous sets, as for loxodromic Möbius recurrences outside an avoided region [1808.09813]. ## 4. Functional equations, integral equations, and nonclassical ambient spaces The classical functional-equation lineage remains central. For continuous generalized additive equations on a homogeneous space \(X=G/L\) of a strongly amenable topological group, if \[ \|f(xy)-f(x)-h(y)\|\le \delta, \] then there exist continuous maps \(F\) and \(H\) satisfying the exact relations \[ H(yz)=H(y)+H(z), \qquad F(xy)=F(x)+H(y), \] together with \[ \|H-h\|\le \delta, \qquad \|F-f\|\le 2\delta \] [1506.06521]. In the discrete Banach-module setting, the same framework accommodates the more general cocycle-type equation \(F(xy)=F(x)\cdot y+H(y)\) [1506.06521]. Integral equations provide a nonlinear analogue. For Volterra equations subject to the approximate inequality \[ \left|y(x)-\int_a^x V(x,\tau,y(\tau))\,d\tau\right|\le \theta, \] a generalized contractive condition \[ |V(x,\tau,y)-V(x,\tau,z)|\le \varphi(|y-z|)\,|y-z| \] yields a unique exact solution \(y_0\) with \[ |y(x)-y_0(x)|\le \frac{\theta}{1-\delta} \] [1503.07967]. Nonlinear dynamics can also exhibit Hyers-Ulam stability only after geometric restrictions are imposed. For loxodromic Möbius recurrences \[ z_{n+1}=g(z_n), \qquad g(z)=\frac{az+b}{cz+d}, \] stability holds for approximate orbits whose initial point lies outside an avoided region \(\mathcal R_g(\infty)\), built from neighborhoods of the backward orbit \(g^{-n}(\infty)\) [1808.09813]. This is a direct reminder that the theory is not uniformly global merely because the defining inequality is global. Recent generalizations extend Hyers-Ulam stability far beyond normed linear spaces. In locally convex cones, the approximate midpoint quadratic equation \[ 2f\left(\frac{x+y}{2}\right)+2f\left(\frac{x-y}{2}\right)\in v(f(x)+f(y))v \] admits a unique quadratic mapping \(Q\) satisfying \[ Q(x)\in (\gamma v)(f(x))(\gamma v) \] under separation and symmetric completeness assumptions [2504.08013]. On weighted graphs, approximate monotonicity, subadditivity, and convexity of the subgraph-weight map can be corrected to exact structural properties on the same vertex and edge set, with explicit bounds such as \[ \|w-\widetilde w\|\le \frac{\varepsilon}{2} \] for the monotone case and \[ \|w-\overline w\|\le \varepsilon \] for the subadditive case [2602.03880]. ## 5. Operator-theoretic characterizations One of the deepest reorganizations of the subject occurs in operator theory. For a closed operator \(T\) between Hilbert spaces, Hyers-Ulam stability is equivalent to closedness of the range and to existence of a bounded Moore-Penrose inverse: \[ T \text{ has HUS } \Longleftrightarrow T^\dagger\in B(Y,X) \Longleftrightarrow R(T)\text{ is closed}. \] Moreover, the best stability constant is \[ K_T=\|T^\dagger\|=\gamma(T)^{-1}, \] where \(\gamma(T)\) is the reduced minimum modulus [1211.0192]. In this formulation, Hyers-Ulam stability becomes a precise inverse estimate rather than a purely qualitative approximation statement. The same philosophy extends to multivalued linear relations. For a closed linear relation \(T\in CR(H,K)\), Hyers-Ulam stability is characterized by \[ T \text{ is HUS } \iff R(T)\text{ is closed}, \] and the optimal constant is \[ M_T=\|T^{-1}\| \] in the quotient sense appropriate to relations [2501.15204]. The stability property is invariant under passage to the regular part \(T_{op}\), to the adjoint \(T^*\), to \(T^*T\), \(TT^*\), and to \(|T|=(T^*T)^{1/2}\); there are also sufficient perturbation theorems for sums and products of stable relations [2501.15204]. These results make clear that, in Hilbert-space linear analysis, Hyers-Ulam stability is not merely analogous to closed-range theory; it is identical to it after the correct formulation. A plausible implication is that operator-theoretic Hyers-Ulam stability is best viewed as a geometric property of the range and of the effective inverse. ## 6. Hyperbolicity, admissibility, and current directions A major current direction connects Hyers-Ulam stability to hyperbolicity. In finite-dimensional nonautonomous dynamics, the discrete equation \[ x_{n+1}=A_nx_n \] is Hyers-Ulam stable with uniqueness if and only if it admits an exponential dichotomy on \(\mathbb Z\); without uniqueness, the correct object is exponential trichotomy, together with an additional bounded-backward-solution condition. For the ODE \[ x'(t)=A(t)x(t), \qquad t\in\mathbb R, \] the corresponding criterion is summable dichotomy or summable trichotomy rather than exponential boundedness [2401.04516]. In random dynamics, tempered exponential dichotomy yields a random shadowing theorem for small nonlinear perturbations, while uniform exponential dichotomy yields a direct Hyers-Ulam stability theorem for the random linear dynamics; the same framework also preserves Lyapunov exponents under small nonlinear perturbations [1909.08707]. Another current direction is norm asymmetry. For semilinear equations \[ x'=A(t)x+f(t,x), \] \((L^p,L^q)\) Hyers-Ulam stability measures the residual in \(L^q\) and the correction in \(L^p\). Under exponential dichotomy of the linear part and a small Lipschitz constant \(c\), the paper proves \((L^p,L^q)\) Hyers-Ulam stability for \[ \infty\ge p>q>1, \] with an explicit constant obtained by Young’s convolution inequality and a contraction argument [2409.14108]. The same work shows that the restriction \(p\ge q\) is substantive rather than formal, by providing an example where stability fails when \(p