The Origin of ℏ: Composition,
Regularity, and Renormalization

IThe Central Claim

"Planck's constant is not an input to physics. It is the structural constant forced by the requirement that composed predictions remain well-defined."

Three companion notes — on refinement compatibility (RCP), the Lipschitz boundary, and the RG-Lipschitz theorem — converge on a single thesis. Planck's constant $\hbar$ is not a free parameter imported from experiment: it is forced into existence by the demand that physical laws survive controlled changes of partition, representation, and scale. The three arguments form a chain of increasing scope, not three independent derivations: composition forces $\hbar$; regularity analysis characterizes the singularity structure at $\hbar=0$; renormalization extends the framework beyond the Kato class where even $\hbar > 0$ is insufficient.

The RCP framework shows that the semigroup composition law for transition kernels has no consistent solution at $\hbar = 0$ or $\hbar = \infty$, forcing a finite, nonzero action-scale constant. The Lipschitz analysis — a regularity property of the kernel produced by composition — shows that $\hbar > 0$ converts singular, delta-supported classical "kernels" into Gaussian-regularized objects with finite spatial Lipschitz bounds, and that the classical limit $\hbar \to 0$ is a Lipschitz catastrophe. The RG-Lipschitz synthesis extends the hierarchy one level further, placing the Lipschitz condition, the Kato class, and full renormalizability into a single chain that tracks how far beyond the classical domain each level of structural completion can reach.

IILevel I — Composition Forces $\hbar$

The most dramatic origin story comes from the partition-compatibility channel of the RCP framework. Consider action-based dynamics on $\mathbb{R}^d$ with a mass parameter $m$, and let $\{K(x,y;t)\}_{t>0}$ be a family of transition kernels. Impose a single physical requirement:

By the Hille–Yosida theorem [HY48], (A1) plus strong continuity forces the existence of an infinitesimal generator — the Hamiltonian is produced, not postulated.

The probabilistic version is Kolmogorov's 1931 theorem [Ko31]: the Chapman–Kolmogorov equation with continuity forces the generator to be a second-order differential operator (for diffusion) or an integral operator (for jumps).

The short-time kernel is forced into the Gaussian form $K(x,y;\varepsilon) \sim \left(\frac{m}{2\pi\hbar\varepsilon}\right)^{d/2} \exp\!\left(\frac{iS_{\mathrm{cl}}}{\hbar}\right)$, and this form is the unique family that closes under the semigroup composition integral.

IIILevel II — $\hbar$ as a Lipschitz Regularizer

The Lipschitz-boundary note analyzes a regularity property of the kernel produced by composition. The key insight: the Lipschitz condition on the force field is the precise mathematical boundary between where classical refinement converges and where it fails.

Classical mechanics lives inside the Lipschitz boundary

The Picard–Lindelöf theorem guarantees existence and uniqueness of solutions to $\dot{y} = f(y)$ when $f$ is Lipschitz continuous with constant $L$, with Gronwall's inequality bounding trajectory separation: $|\Phi_t(x_1) - \Phi_t(x_2)| \le e^{Lt}|x_1 - x_2|$. Without the Lipschitz bound, uniqueness fails — Peano's $\dot{y} = y^{2/3}$ [Pe1890] is the standard counterexample.

The heat kernel's Lipschitz constant

The Euclidean free propagator in $d$ dimensions has spatial Lipschitz constant:

At $\hbar = 0$, the "kernel" collapses to $\delta^{(d)}(x - x_{\mathrm{cl}}(y,T))$ whose Lipschitz constant is infinite. The parameter $\hbar > 0$ converts delta-supported classical kernels into Gaussian-regularized kernels with finite Lipschitz bounds.

Composition improves regularity

A striking fact: semigroup composition is a smoothing operation. For any partition $0 = t_0 < t_1 < \cdots < t_N = T$, the composed kernel's Lipschitz constant equals that of the single kernel at time $T$, regardless of the partition. Each individual short-time slice has Lipschitz constant $\sim (\hbar\,\Delta t_k)^{-(d/2+1)}$, far larger than the composed whole. Composition automatically cancels the excess singularity — but only for $\hbar > 0$.

The paths themselves are not Lipschitz

The path integral is supported on paths that are almost surely Hölder-$\tfrac{1}{2}$ but not Lipschitz — their quadratic variation is $[X]_T = \hbar T/m > 0$. The Cameron–Martin theorem makes this precise: the set of classical (smooth, Lipschitz) paths has Wiener measure zero. The classical limit $\hbar \to 0$ is a concentration phenomenon — the measure collapses toward the classical path — not a restriction to a dominant subset.

IVLevel III — When Even $\hbar$ Is Not Enough

The RG-Lipschitz note extends the story one level further. Even with $\hbar > 0$ providing Lipschitz regularization, there exist potentials singular enough to overwhelm it. The natural domain where $\hbar$-regularization suffices is the Kato class $\mathcal{K}_d$.

The paradigmatic example: the Coulomb potential $V(r) = -Ze^2/r$ in $d = 3$. Its gradient diverges as $r^{-2}$ at the origin, so the force is not Lipschitz — classical mechanics breaks down at head-on collision. But $V \in \mathcal{K}_3$: the Kato–Rellich theorem guarantees a unique self-adjoint Hamiltonian. The hydrogen atom is well-posed quantum-mechanically precisely because $\hbar$ regularizes the Lipschitz singularity.

But the 2D contact interaction $H = -(\hbar^2/2m)\Delta + g\,\delta^{(2)}(x)$ is not in $\mathcal{K}_2$. The 1-loop integral $\int_{|p|\le\Lambda} d^2p/(p^2+M^2) = (1/2\pi)\log((\Lambda^2+M^2)/M^2)$ diverges logarithmically, and even $\hbar$-regularization cannot produce a finite theory. Renormalization enters as a third level of structural completion: the running coupling $g(\Lambda) = g_0/(1 - g_0\log(\Lambda/\Lambda_0)/(2\pi))$ absorbs the divergence, yielding $\beta(g_R) = g_R^2/(2\pi)$.

VThe Three-Level Hierarchy

The hierarchy is strict: $\{\nabla V \text{ Lipschitz}\} \subsetneq \{V \in \mathcal{K}_d\} \subsetneq \{\text{renormalizable}\}.$ The first inclusion is witnessed by the Coulomb potential ($1/r \in \mathcal{K}_3$ but $\nabla(1/r)$ is not Lipschitz). The second inclusion requires that every Kato-class potential be trivially renormalizable (no counterterms needed), which follows from the Kato–Rellich theorem guaranteeing self-adjoint $H$ without renormalization [RS75].

VIThe Classical Limit as Catastrophe

The precise nature of the singularity. The $\hbar\to 0$ limit is essentially singular: the WKB expansion $K \sim \sum a_n\,\hbar^n$ is asymptotic but not convergent, with optimal truncation errors of order $\sim\exp(-S/\hbar)$, and the classical limit is non-injective — distinct quantum systems can share the same asymptotic expansion [Po1886]. The Stokes phenomenon further implies that exponentially small contributions switch on discontinuously at anti-Stokes rays in the complex $\hbar$-plane.

VIISynthesis: The Chain of Forced Completions

VIIIThe Butcher–Hopf Algebra: One Operation at Every Level

A fourth companion note (Trees-Cornerstone) identifies the algebraic backbone: the Connes–Kreimer Hopf algebra $\mathcal{H}_{\mathrm{CK}}$ of rooted trees — the same algebra that John Butcher discovered in 1972 to organize numerical ODE methods.

The derivative is a single counterterm subtraction

Each application of the derivative is one counterterm subtraction. This is the atomic unit of the entire algebraic story.

Trees as organized subtractions

Characters and the exact flow

A character of $\mathcal{H}_{\mathrm{CK}}$ is an algebra map $\phi\!:\mathcal{H}_{\mathrm{CK}}\to\mathbb{R}$. The exact flow defines the character $\phi_{\mathrm{exact}}(\tau) = 1/\sigma(\tau)$:

Composition is the Hopf product

Why $\hbar$ enters: from one character to all characters

Classically, only one character $\phi_{\mathrm{exact}}$ contributes. But composition forces smooth kernels, which require a weighted sum over all paths $K(x,y;t) \sim \int_{\text{paths}} \exp(iS[\text{path}]/\hbar)$. In the tree language, this is the passage from one character to a measure over characters weighted by $e^{iS/\hbar}$. As $\hbar\to 0$, stationary phase collapses it back onto $\phi_{\mathrm{exact}}$ alone.

Renormalization uses the same coproduct

When the sum over characters diverges — as it must beyond the Kato class — renormalization is the Birkhoff decomposition $\phi = \phi_-^{-1}\star\phi_+$ in the Butcher group. The divergent part $\phi_-$ (encoding subdivergences) is extracted using the coproduct $\Delta$. The renormalized character $\phi_+$ is finite.

The tree hierarchy mirrors the regularity hierarchy

Honest gap. The path integral is not literally a sum over B-series characters. A generic Feynman path is Hölder-$\frac{1}{2}$, not $C^\infty$ — it has no B-series expansion. The tree-algebraic description applies to the perturbative expansion of the propagator, not to individual paths. Non-perturbative content (tunneling, instantons, resurgence) goes beyond the tree algebra.

IXFoundational Implications: Measurement as Semigroup Severance

What composition forces: the full arena

The wavefunction is a derived object: given the kernel $K(x,y;t)$ forced by composition, $\psi(x,t) = \int K(x,y;t)\,\psi_0(y)\,d^dy$. Interference is automatic once $\hbar > 0$ forces complex amplitudes: $|K(x,y;t_1+t_2)|^2 \neq \int |K(x,z;t_1)|^2\,|K(z,y;t_2)|^2\,d^dz$. No superposition postulate is needed — it is a theorem of the composition law.

Gleason's theorem [Gl57]: for Hilbert spaces of dimension $\ge 3$, the only frame function (countably additive probability measure on closed subspaces) is the Born rule $p(\phi) = |\langle\psi|\phi\rangle|^2$. The chain (noting that the $\dim\ge 3$ and frame-function assumptions are additional inputs, not derived from (A1)):

Measurement is a local classical limit

A measurement is the assertion that some subsystem — the apparatus — is well-described by the classical limit. The apparatus has $S_{\mathrm{app}}/\hbar \gg 1$: the pointer does not tunnel between positions. Section VI proved that $\hbar \to 0$ is an essential singularity. Measurement is the act of approaching an essential singularity for a specific subsystem.

Semigroup severance: an interpretive framing of collapse

When the apparatus forces the local classical limit at the moment of measurement $t_m$, the transition kernel degenerates to $K(x,y;t) \xrightarrow{S_{\mathrm{app}}/\hbar\,\to\,\infty} \delta^{(d)}\!\big(x - x_{\mathrm{cl}}(y,t)\big)$. The delta function is not an $L^1$ function. The composition integral fails. In composition language, the semigroup timeline is severed.

Decoherence as asymptotic remainder

For a macroscopic apparatus with $S_{\mathrm{app}} \gg \hbar$, the off-diagonal elements of the reduced density matrix decay as $\rho_{\mathrm{off}} \sim \exp\!\left(-S_{\mathrm{app}}/\hbar\right)$ — the non-perturbative terms invisible at every finite order of the classical expansion [JZ85, Zu03]. The pointer basis is selected by the physical apparatus, not by the quantum theory alone.

XConclusion: Differentiability Forces $\hbar$

The thesis: demanding differentiability of the propagation kernel forces $\hbar > 0$.

Smooth kernels are doubly exceptional: topologically (nowhere-differentiable functions are comeager in $C[0,1]$ by the Banach–Mazurkiewicz theorem [BM31]) and measure-theoretically (Wiener-typical paths are differentiable nowhere). The propagation kernel is smooth in its arguments precisely because $\hbar > 0$ Gaussian-regularizes the short-time limit; the paths it sums over are not smooth. The kernel's regularity is the output of composition, not a generic property of function spaces.

The kernel Lipschitz bound $\mathrm{Lip}_x(K_t) \sim \left(\frac{m}{\hbar\, t}\right)^{d/2+1}$ is finite precisely because $\hbar > 0$. One trades regularity of the kernel for roughness of the paths. Differentiability of the kernel forces non-differentiability of the paths.

A final remark. The complex exponential $e^{iS/\hbar}$ is already present in the zero-dimensional static variational problem. The Dirac measure $\delta(f'(x))$ concentrated on critical points of $f$ admits the integral representation $\langle\delta_f \mid g\rangle = \iint \lim_{\varepsilon\to 0} e^{i(f(y)-f(x))/\varepsilon}\; g(x)\;dx\,dy$. The passage from this 0+0 problem to the 0+1 problem (dynamics) requires the regularized measure to converge, which forces a nonzero control parameter $h > 0$ [Ri98].

$\hbar > 0$ is the price of dynamics: the minimal structural constant that makes composed predictions well-defined. Measurement, in the semigroup-severance framing, is the return to statics — the local classical limit where composition fails.

References