The original proof was given by Friedrich Hartogs in 1906, using Cauchy's integral formula for functions of several complex variables.5 Today, usual proofs rely on either the Bochner–Martinelli–Koppelman formula or the solution of the inhomogeneous Cauchy–Riemann equations with compact support. The latter approach is due to Leon Ehrenpreis who initiated it in the paper (Ehrenpreis 1961). Yet another very simple proof of this result was given by Gaetano Fichera in the paper (Fichera 1957), by using his solution of the Dirichlet problem for holomorphic functions of several variables and the related concept of CR-function:6 later he extended the theorem to a certain class of partial differential operators in the paper (Fichera 1983), and his ideas were later further explored by Giuliano Bratti.7 Also the Japanese school of the theory of partial differential operators worked much on this topic, with notable contributions by Akira Kaneko.8 Their approach is to use Ehrenpreis's fundamental principle.
For example, in two variables, consider the interior domain
in the two-dimensional polydisk Δ 2 = { z ∈ C 2 ; | z 1 | < 1 , | z 2 | < 1 } {\displaystyle \Delta ^{2}=\{z\in \mathbb {C} ^{2};|z_{1}|<1,|z_{2}|<1\}} where 0 < ε < 1. {\displaystyle 0<\varepsilon <1.}
Theorem Hartogs (1906): Any holomorphic function f {\displaystyle f} on H ε {\displaystyle H_{\varepsilon }} can be analytically continued to Δ 2 . {\displaystyle \Delta ^{2}.} Namely, there is a holomorphic function F {\displaystyle F} on Δ 2 {\displaystyle \Delta ^{2}} such that F = f {\displaystyle F=f} on H ε . {\displaystyle H_{\varepsilon }.}
Such a phenomenon is called Hartogs's phenomenon, which lead to the notion of this Hartogs's extension theorem and the domain of holomorphy.
Ehrenpreis' proof is based on the existence of smooth bump functions, unique continuation of holomorphic functions, and the Poincaré lemma — the last in the form that for any smooth and compactly supported differential (0,1)-form ω on Cn with ∂ω = 0, there exists a smooth and compactly supported function η on Cn with ∂η = ω. The crucial assumption n ≥ 2 is required for the validity of this Poincaré lemma; if n = 1 then it is generally impossible for η to be compactly supported.10
The ansatz for F is φ f − v for smooth functions φ and v on G; such an expression is meaningful provided that φ is identically equal to zero where f is undefined (namely on K). Furthermore, given any holomorphic function on G which is equal to f on some open set, unique continuation (based on connectedness of G \ K) shows that it is equal to f on all of G \ K.
The holomorphicity of this function is identical to the condition ∂v = f ∂φ. For any smooth function φ, the differential (0,1)-form f ∂φ is ∂-closed. Choosing φ to be a smooth function which is identically equal to zero on K and identically equal to one on the complement of some compact subset L of G, this (0,1)-form additionally has compact support, so that the Poincaré lemma identifies an appropriate v of compact support. This defines F as a holomorphic function on G; it only remains to show (following the above comments) that it coincides with f on some open set.
On the set Cn \ L, v is holomorphic since φ is identically constant. Since it is zero near infinity, unique continuation applies to show that it is identically zero on some open subset of G \ L.11 Thus, on this open subset, F equals f and the existence part of Hartog's theorem is proved. Uniqueness is automatic from unique continuation, based on connectedness of G.
The theorem does not hold when n = 1. To see this, it suffices to consider the function f(z) = z−1, which is clearly holomorphic in C \ {0}, but cannot be continued as a holomorphic function on the whole of C. Therefore, the Hartogs's phenomenon is an elementary phenomenon that highlights the difference between the theory of functions of one and several complex variables.
See the original paper of Hartogs (1906) and its description in various historical surveys by Osgood (1966, pp. 56–59), Severi (1958, pp. 111–115) and Struppa (1988, pp. 132–134). In particular, in this last reference on p. 132, the Author explicitly writes :-"As it is pointed out in the title of (Hartogs 1906), and as the reader shall soon see, the key tool in the proof is the Cauchy integral formula". - Hartogs, Fritz (1906), "Einige Folgerungen aus der Cauchyschen Integralformel bei Funktionen mehrerer Veränderlichen.", Sitzungsberichte der Königlich Bayerischen Akademie der Wissenschaften zu München, Mathematisch-Physikalische Klasse (in German), 36: 223–242, JFM 37.0443.01 https://archive.org/stream/bub_gb_N-sAAAAAYAAJ#page/n229/mode/1up ↩
See for example Vladimirov (1966, p. 153), which refers the reader to the book of Fuks (1963, p. 284) for a proof (however, in the former reference it is incorrectly stated that the proof is on page 324). - Vladimirov, V. S. (1966), Ehrenpreis, L. (ed.), Methods of the theory of functions of several complex variables. With a foreword of N.N. Bogolyubov, Cambridge-London: The M.I.T. Press, pp. XII+353, MR 0201669, Zbl 0125.31904 https://mathscinet.ams.org/mathscinet-getitem?mr=0201669 ↩
See Brown (1936) and Osgood (1929). - Brown, Arthur B. (1936), "On certain analytic continuations and analytic homeomorphisms", Duke Mathematical Journal, 2: 20–28, doi:10.1215/S0012-7094-36-00203-X, JFM 62.0396.02, MR 1545903, Zbl 0013.40701 https://doi.org/10.1215%2FS0012-7094-36-00203-X ↩
See Fichera (1983) and Bratti (1986a) (Bratti 1986b). - Fichera, Gaetano (1983), "Sul fenomeno di Hartogs per gli operatori lineari alle derivate parziali", Rendiconti Dell' Istituto Lombardo di Scienze e Lettere. Scienze Matemàtiche e Applicazioni, Series A. (in Italian), 117: 199–211, MR 0848259, Zbl 0603.35013 https://mathscinet.ams.org/mathscinet-getitem?mr=0848259 ↩
Fichera's proof as well as his epoch making paper (Fichera 1957) seem to have been overlooked by many specialists of the theory of functions of several complex variables: see Range (2002) for the correct attribution of many important theorems in this field. - Fichera, Gaetano (1957), "Caratterizzazione della traccia, sulla frontiera di un campo, di una funzione analitica di più variabili complesse", Rendiconti della Accademia Nazionale dei Lincei, Classe di Scienze Fisiche, Matematiche e Naturali, series 8 (in Italian), 22 (6): 706–715, MR 0093597, Zbl 0106.05202 https://mathscinet.ams.org/mathscinet-getitem?mr=0093597 ↩
See Bratti (1986a) (Bratti 1986b). - Bratti, Giuliano (1986a), "A proposito di un esempio di Fichera relativo al fenomeno di Hartogs" [About an example of Fichera concerning Hartogs's phenomenon], Rendiconti della Accademia Nazionale delle Scienze Detta dei XL, serie 5 (in Italian and English), X (1): 241–246, MR 0879111, Zbl 0646.35007, archived from the original on 2011-07-26 https://web.archive.org/web/20110726235834/http://www.accademiaxl.it/Biblioteca/Pubblicazioni/browser.php?VoceID=2020 ↩
See his paper (Kaneko 1973) and the references therein. - Kaneko, Akira (January 12, 1973), "On continuation of regular solutions of partial differential equations with constant coefficients", Proceedings of the Japan Academy, 49 (1): 17–19, doi:10.3792/pja/1195519488, MR 0412578, Zbl 0265.35008 https://doi.org/10.3792%2Fpja%2F1195519488 ↩
Hörmander 1990, Theorem 2.3.2. - Hörmander, Lars (1990) [1966], An Introduction to Complex Analysis in Several Variables, North–Holland Mathematical Library, vol. 7 (3rd (Revised) ed.), Amsterdam–London–New York–Tokyo: North-Holland, ISBN 0-444-88446-7, MR 1045639, Zbl 0685.32001 https://mathscinet.ams.org/mathscinet-getitem?mr=1045639 ↩
Hörmander 1990, p. 30. - Hörmander, Lars (1990) [1966], An Introduction to Complex Analysis in Several Variables, North–Holland Mathematical Library, vol. 7 (3rd (Revised) ed.), Amsterdam–London–New York–Tokyo: North-Holland, ISBN 0-444-88446-7, MR 1045639, Zbl 0685.32001 https://mathscinet.ams.org/mathscinet-getitem?mr=1045639 ↩
Any connected component of Cn \ L must intersect G \ L in a nonempty open set. To see the nonemptiness, connect an arbitrary point p of Cn \ L to some point of L via a line. The intersection of the line with Cn \ L may have many connected components, but the component containing p gives a continuous path from p into G \ L. ↩