Difference between revisions of "Morse lemma"
Ulf Rehmann (talk | contribs) m (MR/ZBL numbers added) |
|||
Line 1: | Line 1: | ||
− | + | {{MSC|57R45|37}} | |
+ | [[Category:Analysis]] | ||
+ | {{TEX|done}} | ||
− | + | A lemma which describes the structure of the [[Germ|germ]] of a twice continuously-differentiable function. | |
− | + | ===Main statement=== | |
+ | '''Theorem 1''' Let $f:\mathbb R^n\to \mathbb R$ be a function of class $C^\infty$ for which $0$ is a non-degenerate [[Critical point|critical point]], namely $\nabla f (0) =0$ and the Hessian at $0$ has trivial kernel. Then in some neighbourhood $U$ of $0$ there is a local $C^\infty$ coordinate system, namely a $C^\infty$ | ||
+ | diffeomorphism | ||
+ | \[ | ||
+ | \varphi = (x_1, \ldots , x_n): U \to V \subset \mathbb R^n\, , | ||
+ | \] | ||
+ | with $\varphi (0)=0$ and such that the map $\tilde{f} = f\circ \varphi^{-1}$ (namely $\varphi$ in the "$x$-coordinates") takes the form $\tilde{f} (x) = f(0)- x_1^2 - \ldots - x_\lambda^2 + x_{\lambda+1}^2 + \ldots + x_n^2$. | ||
− | + | Here the number $\lambda$ is the [[Morse index|Morse index]] of the critical point $0$ of $f$, that is the number of negative eigenvalues of the Hessian of $f$ at $0$, counted with multiplicities. The assumption $C^\infty$ may be relaxed to $C^p$ for $p\geq 2$, but in this case the change of variables $\varphi$ is in general only of class $C^{p-2}$. If instead $f$ is [[Real analytic function|real analytic]], then $\varphi$ is real analytic. An analogue of the Morse Lemma holds for holomorphic functions of complex variables. | |
− | + | '''Theorem 2''' | |
+ | If $f: \mathbb C^n \to \mathbb C$ is holomorphic in a neighborhood of $0$ and $0$ is a nondegenerate critical point (namely $\frac{\partial f}{\partial z_j} (0) =0$ for every $j$ and the matrix $M_{ij} = \frac{\partial^2 f}{\partial z_i \partial z_j} (0)$ is invertible), then there is a neighborhood $U$ of $0$ with an holomorphic local chart, namely a holomorphic invertible map | ||
+ | \[ | ||
+ | \varphi = (w_1, \ldots , w_n) : \mathbb C^n \supset U\quad \to\quad V\subset \mathbb C^n\, , | ||
+ | \] | ||
+ | such that $\varphi (0) = 0$ and $\tilde{f} = f \circ \varphi^{-1}$ takes the form $f (w) = f(0) + w_1^2 + \ldots + w_n^2$. | ||
− | + | ===Generalizations=== | |
+ | ====Infinite-dimensional case==== | ||
+ | The Morse lemma can be generalized to the infinite-dimensional setting: it holds, for instance, on Banach spaces, see {{Cite|Pa2}}. On separable [[Hilbert space|Hilbert spaces]] it takes the following form. | ||
− | + | '''Theorem 3''' | |
+ | Let $H$ be a separable Hilbert space and $f:H \to \mathbb R$ a $C^k$ function $f$ with $k\geq 3$ (in the sense of [[Frechet differential|Frechet differentiability]]) for which $0$ is a non-degenerate critical point. Then there are convex neighborhoods $U$ and $V$ of $0$, a diffeomorphism (of class $C^{k-2}$) $\varphi: U \to V$ with $\varphi (0)=0$ and a bounded orthogonal projection $P: H \to H$ such that $f (x) = f (0) - \|P (\varphi (x))\|_H^2 + \|\varphi (x) - P (\varphi (x))\|_H^2$. | ||
− | + | Here the dimension of the space ${\rm Im}\, (P)$ coincides with Morse index of the critical point $0$. | |
− | |||
+ | ====Equivariant Morse lemma==== | ||
+ | Consider $f$ as in Theorem 2 and assume that it is invariant with respect to the action of a compact subgroup $G$ of transformations on $\mathbb C^n$. Then the change of variables $\varphi$ can be taken $G$-invariant. An analogous "equivariant Morse lemma" is true in the real-analytic and the differentiable context. Cf. {{Cite|A}} and {{Cite|AGV}}. | ||
+ | ====Morse lemma depending on parameters==== | ||
+ | Let $f: \mathbb R^n \times \R^k \ni (x,\lambda) \mapsto f (x, \lambda)\in \mathbb R$ be a real-valued $C^p$ function with $p\geq 3$. Assume that $D_x f (0,0)=0$ and that $D^2_x f (0,0)$ is nonsingular. Then there exist $C^{k-1}$ coordinates $(z, \lambda)$ in a neighbourhood of $(0,0)$ which brings $f$ to the form | ||
+ | \[ | ||
+ | f (z, \lambda) = f (x(\lambda), \lambda) + \frac{1}{2} \langle A z\, z\rangle\, | ||
+ | \] | ||
+ | where $A$ is a diagonal matrix with entries $\pm 1$ and $\lambda \mapsto x (\lambda)$ is the local (unique) solution of the equations $D_x f (x, \lambda) = 0$ | ||
+ | (the nondegeneracy assumption on $D^2_x f (0,0)$ allows to apply the implicit function theorem and infer that $x$ depends smoothly on $\lambda$). For a proof see {{Cite|H}}, p. 502. | ||
− | === | + | ===References=== |
− | + | {| | |
− | + | |- | |
− | + | |valign="top"|{{Ref|A}}|| V.I. Arnol'd, "Wave front evolution and the equivariant Morse lemma" ''Comm. Pure Appl. Math.'' , '''29''' (1976) pp. 557–582 | |
− | + | |- | |
− | + | |valign="top"|{{Ref|AGV}}|| V.I. Arnol'd, S.M. [S.M. Khusein-Zade] Gusein-Zade, A.N. Varchenko, "Singularities of differentiable maps" , '''1''' , Birkhäuser (1985) {{MR|777682}} {{ZBL|0554.58001}} | |
− | + | |- | |
− | + | |valign="top"|{{Ref|H}}|| L.V. Hörmander, "The analysis of linear partial differential operators" , '''3. Pseudo-differential operators''' , Springer (1985) {{MR|1540773}} {{MR|0781537}} {{MR|0781536}} {{ZBL|0612.35001}} {{ZBL|0601.35001}} | |
− | + | |- | |
− | + | |valign="top"|{{Ref|M}}|| M. Morse, "The calculus of variations in the large" , Amer. Math. Soc. (1934) {{MR|1451874}} {{MR|1501555}} {{MR|1561686}} {{MR|1501489}} {{MR|1501428}} {{ZBL|0011.02802}} {{ZBL|60.0450.01}} | |
− | + | |- | |
− | + | |valign="top"|{{Ref|Pa}}|| R.S. Palais, "Morse theory on Hilbert manifolds", ''Topology'' , '''2''' (1963) pp. 299–340 {{MR|0158410}} {{ZBL|0122.10702}} | |
− | + | |- | |
− | + | |valign="top"|{{Ref|Pa2}}|| R.S. Palais, "The Morse lemma for Banach spaces", "Bull. Amer. Math. Soc.", '''75''' (1969), pp. 968-971. | |
− | + | |- | |
− | + | |} | |
− |
Revision as of 09:30, 28 June 2014
2020 Mathematics Subject Classification: Primary: 57R45 Secondary: 37-XX [MSN][ZBL]
A lemma which describes the structure of the germ of a twice continuously-differentiable function.
Main statement
Theorem 1 Let $f:\mathbb R^n\to \mathbb R$ be a function of class $C^\infty$ for which $0$ is a non-degenerate critical point, namely $\nabla f (0) =0$ and the Hessian at $0$ has trivial kernel. Then in some neighbourhood $U$ of $0$ there is a local $C^\infty$ coordinate system, namely a $C^\infty$ diffeomorphism \[ \varphi = (x_1, \ldots , x_n): U \to V \subset \mathbb R^n\, , \] with $\varphi (0)=0$ and such that the map $\tilde{f} = f\circ \varphi^{-1}$ (namely $\varphi$ in the "$x$-coordinates") takes the form $\tilde{f} (x) = f(0)- x_1^2 - \ldots - x_\lambda^2 + x_{\lambda+1}^2 + \ldots + x_n^2$.
Here the number $\lambda$ is the Morse index of the critical point $0$ of $f$, that is the number of negative eigenvalues of the Hessian of $f$ at $0$, counted with multiplicities. The assumption $C^\infty$ may be relaxed to $C^p$ for $p\geq 2$, but in this case the change of variables $\varphi$ is in general only of class $C^{p-2}$. If instead $f$ is real analytic, then $\varphi$ is real analytic. An analogue of the Morse Lemma holds for holomorphic functions of complex variables.
Theorem 2 If $f: \mathbb C^n \to \mathbb C$ is holomorphic in a neighborhood of $0$ and $0$ is a nondegenerate critical point (namely $\frac{\partial f}{\partial z_j} (0) =0$ for every $j$ and the matrix $M_{ij} = \frac{\partial^2 f}{\partial z_i \partial z_j} (0)$ is invertible), then there is a neighborhood $U$ of $0$ with an holomorphic local chart, namely a holomorphic invertible map \[ \varphi = (w_1, \ldots , w_n) : \mathbb C^n \supset U\quad \to\quad V\subset \mathbb C^n\, , \] such that $\varphi (0) = 0$ and $\tilde{f} = f \circ \varphi^{-1}$ takes the form $f (w) = f(0) + w_1^2 + \ldots + w_n^2$.
Generalizations
Infinite-dimensional case
The Morse lemma can be generalized to the infinite-dimensional setting: it holds, for instance, on Banach spaces, see [Pa2]. On separable Hilbert spaces it takes the following form.
Theorem 3 Let $H$ be a separable Hilbert space and $f:H \to \mathbb R$ a $C^k$ function $f$ with $k\geq 3$ (in the sense of Frechet differentiability) for which $0$ is a non-degenerate critical point. Then there are convex neighborhoods $U$ and $V$ of $0$, a diffeomorphism (of class $C^{k-2}$) $\varphi: U \to V$ with $\varphi (0)=0$ and a bounded orthogonal projection $P: H \to H$ such that $f (x) = f (0) - \|P (\varphi (x))\|_H^2 + \|\varphi (x) - P (\varphi (x))\|_H^2$.
Here the dimension of the space ${\rm Im}\, (P)$ coincides with Morse index of the critical point $0$.
Equivariant Morse lemma
Consider $f$ as in Theorem 2 and assume that it is invariant with respect to the action of a compact subgroup $G$ of transformations on $\mathbb C^n$. Then the change of variables $\varphi$ can be taken $G$-invariant. An analogous "equivariant Morse lemma" is true in the real-analytic and the differentiable context. Cf. [A] and [AGV].
Morse lemma depending on parameters
Let $f: \mathbb R^n \times \R^k \ni (x,\lambda) \mapsto f (x, \lambda)\in \mathbb R$ be a real-valued $C^p$ function with $p\geq 3$. Assume that $D_x f (0,0)=0$ and that $D^2_x f (0,0)$ is nonsingular. Then there exist $C^{k-1}$ coordinates $(z, \lambda)$ in a neighbourhood of $(0,0)$ which brings $f$ to the form \[ f (z, \lambda) = f (x(\lambda), \lambda) + \frac{1}{2} \langle A z\, z\rangle\, \] where $A$ is a diagonal matrix with entries $\pm 1$ and $\lambda \mapsto x (\lambda)$ is the local (unique) solution of the equations $D_x f (x, \lambda) = 0$ (the nondegeneracy assumption on $D^2_x f (0,0)$ allows to apply the implicit function theorem and infer that $x$ depends smoothly on $\lambda$). For a proof see [H], p. 502.
References
[A] | V.I. Arnol'd, "Wave front evolution and the equivariant Morse lemma" Comm. Pure Appl. Math. , 29 (1976) pp. 557–582 |
[AGV] | V.I. Arnol'd, S.M. [S.M. Khusein-Zade] Gusein-Zade, A.N. Varchenko, "Singularities of differentiable maps" , 1 , Birkhäuser (1985) MR777682 Zbl 0554.58001 |
[H] | L.V. Hörmander, "The analysis of linear partial differential operators" , 3. Pseudo-differential operators , Springer (1985) MR1540773 MR0781537 MR0781536 Zbl 0612.35001 Zbl 0601.35001 |
[M] | M. Morse, "The calculus of variations in the large" , Amer. Math. Soc. (1934) MR1451874 MR1501555 MR1561686 MR1501489 MR1501428 Zbl 0011.02802 Zbl 60.0450.01 |
[Pa] | R.S. Palais, "Morse theory on Hilbert manifolds", Topology , 2 (1963) pp. 299–340 MR0158410 Zbl 0122.10702 |
[Pa2] | R.S. Palais, "The Morse lemma for Banach spaces", "Bull. Amer. Math. Soc.", 75 (1969), pp. 968-971. |
Morse lemma. Encyclopedia of Mathematics. URL: http://encyclopediaofmath.org/index.php?title=Morse_lemma&oldid=24510