PRH | Essay | 7.11 • Complex Analysis as Blur in Disguise

Complex Analysis as Blur in Disguise Poisson, Cauchy, and Epistemic Invariants Aleksandar Perišić November 2025 Abstract The blur method is a meta–framework for doing mathematics under explicitly limited resolution: one chooses a positive averaging kernel and a budget, and then insists that only quantities stable under such blurs are meaningful at that budget. In this note we argue that much of classical complex analysis—Poisson kernels, Cauchy integrals, residues—is already an instance of blur in disguise. The Poisson kernel on the real line is a genuine blur: a positive, normalized approximate identity. Its harmonic extension together with the Cauchy–Riemann equations produces a canonical complex function whose real and imaginary parts are a pair of conjugate blur– invariants of the boundary data. The Cauchy kernel, decomposed into Poisson and Hilbert parts, is the complex analogue of a blur kernel: contour integrals with this kernel collapse all interior microstructure down to a finite list of invariants such as values, derivatives, and residues. From this viewpoint, poles are not places where “complex analysis knows everything”; they are cores of inaccessibility whose detailed behavior is deliberately blurred, leaving behind only a small number of invariants that the theory chooses to remember. Complex analysis is exact about these invariants, but it quietly treats everything else as epistemic blur. 1 Blur, in one paragraph We briefly recall the blur philosophy in a form tailored to harmonic and complex analysis; see [3,4] for a more general discussion. Ablur kernel on Ris a family (kτ)τ>0of nonnegative functions with ZR kτ(x)dx = 1 and kτ→δ0as τ↓0 in the sense of distributions. Blurring a function fat scale τmeans forming the convolution (Bτf)(x) = (kτ∗f)(x) = ZR kτ(x−u)f(u)du. A quantity Q ( f )is blur–invariant at a given budget if we obtain essentially the same value when fis replaced by Bτffor all admissible scales τin the budget. Blur is thus a discipline of explicitly separating what we decide to know from what we tolerate as unresolved. The key move is always the same: choose a blur; track quantities that survive it unchanged; declare everything else invisible to the current theory. 2 The Poisson kernel as a genuine blur We start from the most classical object: the Poisson kernel on the upper half–plane. It is a perfect example of a blur kernel in the analytic sense. 1 2.1 Poisson kernel and harmonic extension Let H={z=x+iy ∈C:y > 0}be the upper half–plane. The Poisson kernel is Py(x) = 1 π y x2+y2, x ∈R, y > 0. For each fixed y > 0, this is a positive function on Rwith Z∞ −∞ Py(x)dx = 1, Py(x)→δ0as y↓0. Thus (Py)y>0is an honest blur kernel on the line, with yplaying the role of resolution. Given a bounded (or suitable) function φ:R→R, its Poisson extension u(x, y) = (Py∗φ)(x) = Z∞ −∞ Py(x−t)φ(t)dt is harmonic on Hand converges nontangentially to φalmost everywhere as y↓0[2]. In blur terms: •The boundary function φis the raw observable. •For each y > 0,u(·, y)is φblurred at scale y. •As ydecreases, the blur radius shrinks, and one recovers φin the limit. So the Poisson extension is literally a blur map from boundary data on R to harmonic data in the half–plane. 2.2 Poisson via Abel transform and Fourier There is also a complementary way to see where the Poisson kernel comes from that fits naturally with blur. Instead of starting from the Dirichlet problem, we start from spherical averaging and then look at its Fourier profile. Very informally, the Abel transform of a function f in higher dimension takes spherical means around a point: one averages f over spheres of radius t centered at that point. If we encode these spherical averages in a kernel K ( t, ξ ), with t > 0playing the role of radius and ξ the frequency variable, then taking a Fourier transform in the spatial variable produces a damping factor of the form e−2πt|ξ|. On the line, this can be written as P(t, x) = FξK(t, ·)(x) = ZR e−2πt|ξ|e−2πiξx dξ, which is exactly the Poisson kernel (up to the usual normalization and identifying t with y ). In other words: • The factor e−2πt|ξ| is a frequency–space blur: it exponentially damps high frequencies at rate proportional to t. • Taking the inverse Fourier transform in ξ converts this frequency blur into the spatial blur kernel Pt(x). • Conceptually, we are looking at a function through a spherical cloud of radius t around each point, and the Poisson kernel tells us how that cloud weights contributions from different boundary locations. 2 Thus Poisson blur can be viewed in two equivalent ways: •as the unique kernel that solves the Dirichlet problem in the upper half–plane, and • as the spatial avatar of an exponential frequency blur that arises from spherical averaging (Abel transform) followed by a Fourier transform. In both views, ( Pt ) t>0 is the canonical way to blur a function at scale t : it is the blur profile seen by an observer at height twhose vision is governed by Laplace’s equation. 2.3 Harmonic conjugates and the Hilbert transform If u is harmonic on H with suitable growth bounds, there exists a harmonic function v (unique up to an additive constant) such that f(z) = u(x, y)+iv(x, y) is holomorphic on H and u, v satisfy the Cauchy–Riemann equations [ 1 ]. As y↓ 0, the imaginary part von Hconverges (in a suitable sense) to the Hilbert transform of φon R: (Hφ)(x) = 1 πp.v. Z∞ −∞ φ(t) x−tdt. Thus, starting from real boundary data φ , the blur/extension procedure naturally produces a complex function F(z) = u(x, y)+iv(x, y), with ℜF(x) = φ(x),ℑF(x) = (Hφ)(x) on the boundary. From the blur viewpoint, this is an important pattern: •The real part uis obtained by a positive blur of φwith the Poisson kernel. • The imaginary part v is the unavoidable conjugate component forced by analyticity, reconstructed from φvia a singular integral blur (the Hilbert transform). • Together they form the minimal complex packaging of what can be stably inferred from φ under the Poisson blur. In other words, holomorphic functions on H are exactly what one gets by taking real data on the boundary, blurring it with Poisson, and insisting that the result extends analytically. The imaginary part is not an extra decoration; it is the shadow demanded by the blur and the Cauchy–Riemann equations. 3 The Cauchy kernel as complex blur We now turn to the Cauchy integral formula. It can be read as a complex version of blur: a convolution with a kernel that erases almost all interior detail, leaving only a small set of invariants. 3 3.1 Cauchy integral as boundary blur Let f be holomorphic in a domain containing a simple closed contour γ and its interior. Cauchy’s integral formula says f(z0) = 1 2πi Iγ f(z) z−z0 dz, z0inside γ. This has exactly the structure of a blur: •The kernel Kz0(z) = 1 z−z0 is a complex kernel on the boundary. •It is normalized in the sense that 1 2πi Iγ 1 z−z0 dz = 1. • The value f ( z0 )is read off by integrating f against this kernel along γ ; the interior of γ is never directly inspected. This is strongly reminiscent of a blur operator ( Bf )( x0 ) = Rk ( x0−x ) f ( x ) dx with k positive and Rk = 1. The difference is that Kz0 is complex–valued and supported on a curve instead of on the full line, but the logic is the same: average against a kernel and ignore interior microstructure, trusting that the invariants you care about are preserved. 3.2 Poisson and Hilbert hiding inside Cauchy On the upper half–plane it is convenient to use the Cauchy transform F(z) = 1 πi Z∞ −∞ φ(t) t−zdt, z =x+iy ∈H. Write the kernel explicitly: 1 t−z=1 (t−x)−iy =(t−x) + iy (t−x)2+y2, so that 1 πi 1 t−z=1 π y (t−x)2+y2−i π t−x (t−x)2+y2. The real part of this kernel is exactly the Poisson kernel Py(x−t) = 1 π y (x−t)2+y2, and the imaginary part is (up to sign) the conjugate Poisson kernel that generates the Hilbert transform. Consequently, ℜF(x+iy) = Z∞ −∞ Py(x−t)φ(t)dt = (Py∗φ)(x), while ℑF(x+iy)=−1 πZ∞ −∞ t−x (t−x)2+y2φ(t)dt, which converges (as y↓ 0) to the Hilbert transform Hφ ( x )(up to the usual constant/sign convention). Thus, for this standard normalization of the Cauchy transform: 4 • the Poisson kernel appears in the real part of the Cauchy kernel and produces the genuine blur (harmonic extension); • the Hilbert kernel appears in the imaginary part and produces the conjugate component forced by analyticity. Complex Cauchy integration is therefore precisely a combination of a positive blur (Poisson) and its conjugate (Hilbert) packaged into a single complex kernel. 3.3 Contour deformation as blur–invariance A hallmark of Cauchy theory is that contour integrals of holomorphic functions are invariant under smooth deformations of the contour that do not cross singularities: Iγ1 f(z)dz =Iγ2 f(z)dz whenever γ1and γ2are homotopic in a domain where fis holomorphic. This is a geometric form of blur–invariance: we can wiggle the contour within a given class, and the observable Hf does not change. Only topological data (which singularities are enclosed) matter. Everything else—the precise shape of the path, the detailed behavior of f between singularities—is blurred away. 4 Singularities as blur cores From the blur perspective, singularities are not places where complex analysis “blows up” in an uninteresting way; they are cores of inaccessibility whose detailed microstructure is intentionally ignored once a small number of invariants has been extracted. 4.1 Isolated singularities and Laurent expansions Suppose f has an isolated singularity at a and is holomorphic on 0 <|z−a|< r . Its Laurent expansion reads f(z) = ∞ X k=−m ak(z−a)k. For each integer n≥0we have the coefficient extraction formula an=1 2πi I|z−a|=ε f(z) (z−a)n+1 dz, and for n=−1the residue: Res(f, a) = a−1=1 2πi I|z−a|=ε f(z)dz. In principle, if we are willing to vary the kernel ( z−a ) −n−1 we can recover all coefficients an , both positive and negative. But many of the central results in complex analysis use only a tiny fraction of this information: •The residue theorem depends only on a−1at each pole. •The argument principle counts zeros and poles using only Hf′(z)/f(z)dz. • Large parts of the theory treat all poles of a given order as belonging to one category, regardless of the precise higher coefficients. 5 From a blur standpoint: • A pole is a region where the function can “arrive” in infinitely many different ways (infinitely many possible principal parts). • The contour integral with kernel 1(for residues) or kernels tailored to a few derivatives sees only one or a few Laurent coefficients. • All singular germs with the same extracted coefficients are indistinguishable to those observables; their remaining structure is epistemic blur relative to this theory. Complex analysis is completely exact about the invariants it chooses to track (residues, orders of poles, etc.), but it implicitly declares that no further local data near a singularity will be used and therefore allows all such data to be blurred away. 4.2 Essential singularities as maximal epistemic randomness At an essential singularity, Picard’s theorem says that f takes almost all complex values arbitrarily close to the singularity. There is no finite invariant like “order of pole” that summarizes the behavior; the local image is as wild as analyticity allows. For blur this is a canonical example of maximal epistemic randomness: • Given only contour integrals and residues, there is no compact summary of the behavior near an essential point. • The best the theory can do is to assert that “everything not forbidden by global constraints happens” arbitrarily close to that point. The residue calculus quietly accepts that no meaningful blur–invariant beyond this rough statement exists at essential singularities. They are genuine black boxes for Cauchy–type observables. 4.3 Gauss–type laws and finite invariants The residue theorem Iγ f(z)dz = 2πi X ak∈int(γ) Res(f, ak) has the same structure as Gauss’s divergence theorem in vector calculus: ZZ∂Ω  E·d S=(total charge in Ω). In both cases: •The interior can be arbitrarily complicated. • A boundary integral is completely determined by a finite list of invariants (charges or residues). •Everything else is invisible to that class of observables and may be treated as blur. Complex analysis differs in that analyticity makes this mechanism astonishingly rigid: knowing the boundary values on a curve can determine the function everywhere inside. But the principle is the same: we commit to a small set of invariants and happily collapse all other degrees of freedom into epistemic blur. 6 5 Complex numbers as a blur envelope The constructions above suggest a conceptual reading of complex numbers themselves as the minimal blur envelope over the real line. 5.1 From real data to analytic functions Starting from a real function φ on R , the Poisson blur and the Cauchy–Riemann equations produce a complex function F(z) = u(x, y)+iv(x, y) on H whose real part extends φ and whose imaginary part is forced by blur and analyticity (via the Hilbert transform). Thus: •The real axis carries the original observable φ. • The upper half–plane H carries all holomorphic functions whose boundary real part matches φalmost everywhere. • Each such germ F packages two blur–invariants of φ : its Poisson blur and its Hilbert– conjugate blur. Complex numbers appear here as the coordinates of the smallest space in which these two pieces can be stored at once. The operation “pass from φ to F ” is a canonical blur–based complexification. 5.2 Analytic signals and positive frequencies In signal processing one often passes from a real signal fto its analytic signal fa(x)=f(x) + i(Hf)(x), where Hf is the Hilbert transform. The Fourier transform of fa has support only on nonnegative frequencies [2]: c fa(ξ)=2b f(ξ) (ξ > 0),c fa(ξ) = 0 (ξ < 0). Interpreted through blur: •We start with a real observable f. •We blur/project the Fourier data onto positive frequencies. • The price of this directional blur is that we must keep the conjugate component Hf in the imaginary part to reconstruct f. • The analytic signal fa is the minimal complex object that retains all information compatible with this blur choice. Again, complex numbers serve as a two–coordinate container for an observable and its blur–conjugate, standing in exactly the same relationship as (u, v)in the Poisson picture. 7 5.3 Complex analysis as exact blur calculus Seen from this angle, classical complex analysis can be summarized as follows: •Choose blur kernels (Poisson and Cauchy) adapted to the geometry of the plane. • Restrict attention to the extremely rigid class of holomorphic functions, for which blur and analyticity interact perfectly. • Use contour integrals with these kernels to read off a small set of invariants (values, derivatives, residues) and deliberately treat everything else as irrelevant. The theory is exact and deterministic about these invariants, but it is not omniscient about the functions themselves. It simply chooses a strong blur scheme that collapses infinitely many local degrees of freedom into a finite list of blur–invariants and then builds an exact calculus on top of that list. 6 Conclusion From the blur perspective, complex analysis is not a separate world but a particularly successful specialization of the same underlying philosophy: • The Poisson kernel is a genuine blur: positive, normalized, and forming an approximate identity. Its harmonic extension is precisely a blurred version of boundary data, with the blur radius given by the height yin the half–plane. • The Hilbert transform and harmonic conjugation add the inevitable shadow component demanded by analyticity; together with Poisson they package boundary data into holomorphic functions. • The Cauchy kernel is a complex blur kernel: integrating against it collapses the interior of a contour to a single value or a small family of coefficients, ignoring all further microstructure. • Poles and residues are prototypes of blur cores and blur–invariants: infinitely many distinct local behaviors near a singularity collapse to the same residue for the purposes of the theory. Essential singularities are points where no such compact invariant exists, and the local behavior is maximally blurred. • Contour deformation invariance is the geometric expression of blur–invariance: any two contours enclosing the same singularities yield the same observables, regardless of inner details. In this light, complex analysis appears as an “exact blur calculus”: it chooses extremely rigid objects (holomorphic functions) and extremely structured blur kernels (Poisson and Cauchy), and then reads off exactly those invariants that survive under the allowed blurs. The epistemic step—accepting that everything else is blur—is present, but traditionally left unnamed. The blur framework simply makes that step explicit and shows that the Poisson and Cauchy tools we already use are blur mechanisms in disguise. References [1] L. V. Ahlfors, Complex Analysis, 3rd ed., McGraw–Hill, 1979. [2] E. M. Stein and R. Shakarchi, Fourier Analysis: An Introduction, Princeton University Press, 2003. 8 [3] A. Perišić, Blur as a Universal Principle: Number Theory, Probability, Dynamics, Zenodo, 2025. [4] A. Perišić, Epistemological Blur, Zenodo, 2025. 9