Difference between revisions of "Bernstein method"
|  (Importing text file) | Ulf Rehmann (talk | contribs)  m (tex encoded by computer) | ||
| Line 1: | Line 1: | ||
| + | <!-- | ||
| + | b0157201.png | ||
| + | $#A+1 = 85 n = 0 | ||
| + | $#C+1 = 85 : ~/encyclopedia/old_files/data/B015/B.0105720 Bernstein method, | ||
| + | Automatically converted into TeX, above some diagnostics. | ||
| + | Please remove this comment and the {{TEX|auto}} line below, | ||
| + | if TeX found to be correct. | ||
| + | --> | ||
| + | |||
| + | {{TEX|auto}} | ||
| + | {{TEX|done}} | ||
| + | |||
| ''method of auxiliary functions'' | ''method of auxiliary functions'' | ||
| Line 5: | Line 17: | ||
| A simple example of the application of Bernstein's method is the a priori estimate of the modulus of the derivatives of the solution of the Dirichlet problem for the non-linear (quasi-linear) elliptic equation | A simple example of the application of Bernstein's method is the a priori estimate of the modulus of the derivatives of the solution of the Dirichlet problem for the non-linear (quasi-linear) elliptic equation | ||
| − | + | $$ \tag{* } | |
| + | \left . | ||
| + | |||
| + | \begin{array}{r} | ||
| + | { | ||
| + | \frac{\partial   ^ {2} z }{\partial  x  ^ {2} } | ||
| + |  + | ||
| + | |||
| + | \frac{\partial   ^ {2} z }{\partial  y  ^ {2} } | ||
| + |   =  f | ||
| + | \left ( x, y, z,  | ||
| + | \frac{\partial z}{\partial x} | ||
| + |  ,  | ||
| + | \frac{\partial z}{\partial y} | ||
| + |  \right )  \equiv  }  \\ | ||
| + | {\equiv  a \left (   | ||
| + | \frac{\partial z}{\partial x} | ||
| + |  \right )  ^ {2} + 2b | ||
| + | |||
| + | \frac{\partial z}{\partial x} | ||
| + | |||
| + | \frac{\partial z}{\partial y} | ||
| + |  + c | ||
| + | \left (  | ||
| + | \frac{\partial z}{\partial y} | ||
| + |  \right )  ^ {2} + }  \\ | ||
| + | {+ 2d  | ||
| + | \frac{\partial z}{\partial x} | ||
| + |  + 2e  | ||
| + | \frac{\partial z}{\partial y} | ||
| + |  + | ||
| + | g \   \mathop{\rm in}  \textrm{ the }  \textrm{ disc }  D, }  \\ | ||
| + | \end{array} | ||
| + |   \right \} | ||
| + | $$ | ||
| − | + | $$  | |
| + | z \mid  _ {C}  =  0, | ||
| + | $$ | ||
| − | where  | + | where  $  a, b, c, d, e, g $ | 
| + | are smooth functions of  $  x, y, z $;   | ||
| + | $  C $ | ||
| + | is the circle, the boundary of the disc  $  D $ | ||
| + | with radius  $  R $( | ||
| + | the assumption to the effect that  $  D $ | ||
| + | is a disc and  $  z\mid  _ {C} = 0 $ | ||
| + | is immaterial, since the general case of an arbitrary-connected domain and inhomogeneous boundary condition is readily reduced to the case under consideration by a change of the function and a conformal transformation of the domain). | ||
| − | If  | + | If  $  f _ {z} ^ { \prime } \geq  0 $,   | 
| + | then the estimated maximum modulus  $  n $ | ||
| − | + | $$  | |
| + | n  =  \max _ {(x,y) \in D + C } \  | ||
| + | | z (x, y) | | ||
| + | $$ | ||
| of the solution of problem (*) is immediately obtained from the maximum principle. | of the solution of problem (*) is immediately obtained from the maximum principle. | ||
| − | In order to prove that a regular solution of problem (*) exists it is sufficient to have a priori estimates of the maximum modulus of the derivatives of the solution up to the third order (cf. [[Continuation method (to a parametrized family)|Continuation method (to a parametrized family)]]). To estimate  | + | In order to prove that a regular solution of problem (*) exists it is sufficient to have a priori estimates of the maximum modulus of the derivatives of the solution up to the third order (cf. [[Continuation method (to a parametrized family)|Continuation method (to a parametrized family)]]). To estimate  $  \max _ {C}  | \partial  z/ \partial  x | $ | 
| + | and  $  \max _ {C}  | \partial  z/ \partial  y | $,   | ||
| + | it is sufficient to estimate  $  \max _ {C}  | \partial  z/ \partial  \rho | $( | ||
| + | since  $  z \mid  _ {C} = 0 $),   | ||
| + | where  $  \rho , \theta $ | ||
| + | are polar coordinates in the disc  $  D $.   | ||
| + | Now introduce a new (auxiliary) function  $  u $,  | ||
| + | given by the formula | ||
| + | |||
| + | $$  | ||
| + | z  =  \phi _ {1} (u)  =  -n- \alpha + \alpha   \mathop{\rm ln}  u , | ||
| + | $$ | ||
| + | |||
| + | where  $  \alpha > 0 $ | ||
| + | will be selected later. The function  $  u = u(x, y) $ | ||
| + | varies from  $  e $ | ||
| + | to  $  e ^ {(2n + \alpha )/ \alpha } $ | ||
| + | in the same direction as  $  z(x, y) $( | ||
| + | $  -n \leq  z \leq  n $).   | ||
| + | Since | ||
| + | |||
| + | $$  | ||
| + | |||
| + | \frac{\partial z}{\partial x} | ||
| + |   =   | ||
| + | \frac \alpha {u} | ||
| + | |||
| + | \frac{\partial u}{\partial x} | ||
| + |  , | ||
| + | $$ | ||
| + | |||
| + | $$  | ||
| + | |||
| + | \frac{\partial   ^ {2} z }{\partial x  ^ {2} } | ||
| + |   = \  | ||
| + | |||
| + | \frac \alpha {u} | ||
| + | |||
| + | \frac{\partial   ^ {2} u }{\partial x  ^ {2} } | ||
| + |  - | ||
| + | |||
| + | \frac \alpha {u  ^ {2} } | ||
| + |  \left (  | ||
| + | \frac{\partial u}{\partial x} | ||
| + |  \right )  ^ {2} , | ||
| + | $$ | ||
| + | |||
| + | and similar results for derivatives with respect to  $  y $,   | ||
| + | it follows that  $  u $ | ||
| + | satisfies the equation | ||
| − | + | $$  | |
| − | + | \frac{\partial   ^ {2} u }{\partial  x  ^ {2} } | |
| + |  + | ||
| − | + | \frac{\partial   ^ {2} u }{\partial  y  ^ {2} } | |
| + |   = \  | ||
| − | + | \frac{1}{u} | |
| − | + | \left [  \left (  | |
| + | \frac{\partial u}{\partial x} | ||
| + |  \right )  ^ {2} + | ||
| + | \left (  | ||
| + | \frac{\partial u}{\partial y} | ||
| + |  \right )  ^ {2} \  | ||
| + | \right ] + | ||
| + | $$ | ||
| − | + | $$  | |
| + | +  | ||
| − | + | \frac \alpha {u} | |
| + |  \left [ a \left (  | ||
| + | \frac{\partial u}{\partial x} | ||
| + |  \right )  ^ {2} + 2b  | ||
| + | \frac{\partial u}{\partial x} | ||
| − | + | \frac{\partial u}{\partial y} | |
| + |  + c | ||
| + | \left (  | ||
| + | \frac{\partial u}{\partial y} | ||
| + |  \right )  ^ {2}  \right ] + | ||
| + | $$ | ||
| − | + | $$  | |
| + | +  | ||
| + | 2d  | ||
| + | \frac{\partial u}{\partial x} | ||
| + |  + 2e  | ||
| + | \frac{\partial u}{\partial y} | ||
| + |  + g  | ||
| + | \frac{u} \alpha  | ||
| + |   \equiv  Q. | ||
| + | $$ | ||
| − | + | Let  $  M $ | |
| + | be the upper bound of  $  | a |, | b |, | c | $ | ||
| + | in  $  D $,  | ||
| + | and let  $  \alpha = 1/8 M $.   | ||
| + | If  $  \partial  u/ \partial  x $ | ||
| + | and  $  \partial  u/ \partial  y $ | ||
| + | are considered as current coordinates in the plane, and  $  x, y, z $ | ||
| + | as parameters, the equation  $  Q = 0 $ | ||
| + | is the equation of an ellipse, since the determinant  $  a _ {1} c _ {1} - b _ {1}  ^ {2} > 2 u  ^ {2} /3 $,  | ||
| + | where | ||
| − | + | $$  | |
| + | a _ {1}  =   | ||
| + | \frac{1}{u} | ||
| + |  \left ( 1+  | ||
| + | \frac{a}{8M} | ||
| + |  \right ) ,\ \  | ||
| + | b _ {1}  =   | ||
| + | \frac{b}{8Mu} | ||
| + |  ,\ \  | ||
| + | c _ {1}  =   | ||
| + | \frac{1}{u} | ||
| − | + | \left ( 1 +  | |
| + | \frac{t}{8M} | ||
| + |  \right ) . | ||
| + | $$ | ||
| + | |||
| + | Thus, for any  $  \partial  u/ \partial  x $ | ||
| + | and  $  \partial  u/ \partial  y $,  | ||
| + | $  Q $ | ||
| + | will not be smaller than a certain negative number  $  -P $,  | ||
| + | $  Q \geq  -P $( | ||
| + | the number  $  P $ | ||
| + | is readily obtained in explicit form). If one introduces the function  $  u _ {1} $ | ||
| + | given by the formula | ||
| + | |||
| + | $$  | ||
| + | u _ {1}  =  u +  | ||
| + | \frac{P}{4} | ||
| + |  (x  ^ {2} +y  ^ {2} ), | ||
| + | $$ | ||
| one obtains | one obtains | ||
| − | + | $$  | |
| − | + | \frac{\partial   ^ {2} u _ {1} }{\partial  x  ^ {2} } | |
| − | + | +  | |
| + | \frac{\partial   ^ {2} u _ {1} }{\partial y  ^ {2} } | ||
| − | + |  =  Q + P  \geq   0, | |
| + | $$ | ||
| − | + | and  $  u _ {1} $ | |
| + | attains its maximum on the boundary  $  C $ | ||
| + | of the domain  $  D $ | ||
| + | and, since  $  u _ {1} $ | ||
| + | is constant on  $  C $,  | ||
| + | one has | ||
| − | + | $$  | |
| − | + | \frac{\partial  u _ {1} }{\partial  \rho } | |
| + |   \geq   0 \ \  | ||
| + | \textrm{ and } \ \  | ||
| + | |||
| + | \frac{\partial u}{\partial \rho} | ||
| + |   \geq  \  | ||
| + | -  | ||
| + | \frac{1}{2} | ||
| + |  PR , | ||
| + | $$ | ||
| + | |||
| + | where  $  R $ | ||
| + | is the radius of the circle  $  C $.  | ||
| + | Hence it is possible to find a negative lower bound for  $  \partial  z/ \partial  \rho $: | ||
| + | |||
| + | $$  | ||
| + | |||
| + | \frac{\partial z}{\partial \rho} | ||
| + |   = \  | ||
| + | |||
| + | \frac \alpha {u} | ||
| + | |||
| + | \frac{\partial u}{\partial \rho} | ||
| + |   \geq  \  | ||
| + | -   | ||
| + | \frac{\alpha PR }{2e ^ {(2n + \alpha )/ \alpha } } | ||
| + |  . | ||
| + | $$ | ||
| + | |||
| + | If the same reasoning is applied to a second auxiliary function  $  u $ | ||
| + | |||
| + | $$  | ||
| + | z  =  \phi _ {2} (u)  = \  | ||
| + | -n- \alpha + \alpha | ||
| + |   \mathop{\rm ln}   | ||
| + | \frac{1}{1-u} | ||
| + |  , | ||
| + | $$ | ||
| one obtains an estimate from above | one obtains an estimate from above | ||
| − | + | $$  | |
| + | |||
| + | \frac{\partial z}{\partial \rho} | ||
| + |   \leq  \  | ||
| + | |||
| + | \frac{\alpha P _ {1} R }{2} | ||
| + | |||
| + | e ^ {(n+ \alpha )/2 } . | ||
| + | $$ | ||
| + | |||
| + | Thus,  $  \max _ {C}  | \partial  z/ \partial  \rho | $ | ||
| + | is estimated, which means that  $  \max _ {C}  | \partial  z/ \partial  x | $ | ||
| + | and  $  \max _ {C}  | \partial  z/ \partial  y | $ | ||
| + | are estimated as well. The estimate of the maximum modulus of the first derivatives inside the domain  $  D $ | ||
| + | is performed in a similar manner: introduce an auxiliary function  $  u $ | ||
| + | given by the formula | ||
| + | |||
| + | $$  | ||
| + | z  =  \phi _ {3} (u)  =  -n + \alpha \  | ||
| + |  \mathop{\rm ln}   \mathop{\rm ln}  u . | ||
| + | $$ | ||
| + | |||
| + | The function  $  u $ | ||
| + | varies in the same direction as  $  z $,  | ||
| + | from  $  e $ | ||
| + | to  $  e ^ {e ^ {2n/ \alpha } } $.   | ||
| + | In view of (*), on may write the following expression for  $  u $ | ||
| + | |||
| + | $$  | ||
| + | |||
| + | \frac{\partial   ^ {2} u }{\partial x  ^ {2} } | ||
| + | |||
| + | + | ||
| + | |||
| + | \frac{\partial   ^ {2} u }{\partial y  ^ {2} } | ||
| + |   = \  | ||
| + | |||
| + | \frac{1}{u   \mathop{\rm ln}  u } | ||
| + | |||
| + | \left [ | ||
| + | (1+  \mathop{\rm ln}  u + \alpha a) | ||
| + | \left (  | ||
| + | \frac{\partial u}{\partial x} | ||
| + |  \right )  ^ {2} +2ab | ||
| + | |||
| + | \frac{\partial u}{\partial x} | ||
| − | + | \frac{\partial u}{\partial y\right} | |
| + |  {} + | ||
| + | $$ | ||
| − | + | $$  | |
| + | + \left . | ||
| + | (1 +  \mathop{\rm ln}  u + \alpha c) \left (  | ||
| + | \frac{\partial u}{\partial y} | ||
| + |  \right )  ^ {2}  \right ] + 2d | ||
| − | + | \frac{\partial u}{\partial x} | |
| + |  + 2e | ||
| − | + | \frac{\partial u}{\partial y} | |
| + |  + g | ||
| − | + | \frac{u   \mathop{\rm ln}  u } \alpha  | |
| + |   \equiv  Q _ {1} . | ||
| + | $$ | ||
| Considerations similar to those given above show that if the function | Considerations similar to those given above show that if the function | ||
| − | + | $$  | |
| + | w  =  \left (  | ||
| + | \frac{\partial u}{\partial x} | ||
| + |  \right )  ^ {2} + | ||
| + | \left (  | ||
| + | \frac{\partial u}{\partial y} | ||
| + |  \right )  ^ {2} | ||
| + | $$ | ||
| − | attains a maximum in the domain  | + | attains a maximum in the domain  $  D $,   | 
| + | this maximum does not exceed some number, the value of which depends solely on  $  n $ | ||
| + | and  $  M $.   | ||
| + | This yields the required estimates of  $  \max _ {D}  | \partial  z/ \partial  x | $ | ||
| + | and  $  \max _ {D}  | \partial  z/ \partial  y | $. | ||
| − | Bernstein's method may also be used to estimate, in a similar manner, the maximum modulus in the domain  | + | Bernstein's method may also be used to estimate, in a similar manner, the maximum modulus in the domain  $  D + C $ | 
| + | of all highest derivatives of the solution (the only other operation which is required is the differentiation of the initial equation). | ||
| The method was first utilized by S.N. Bernstein [e theory of functions','../c/c025430.htm','Continuation method (to a parametrized family)','../c/c025520.htm','Euler–Lagrange equation','../e/e036510.htm','Fourier series','../f/f041090.htm','Functions of a real variable, theory of','../f/f042130.htm','Hilbert problems','../h/h120080.htm','Jackson inequality','../j/j054000.htm','Laplace theorem','../l/l057530.htm','Lebesgue constants','../l/l057800.htm','Limit theorems','../l/l058920.htm','Linear elliptic partial differential equation and system','../l/l059180.htm','Lyapunov theorem','../l/l061200.htm','Mathematical statistics','../m/m062710.htm','Minimal surface','../m/m063920.htm','Markov–Bernstein-type inequalities','../m/m110060.htm','Ornstein–Uhlenbeck process','../o/o070240.htm','Orthogonal polynomials','../o/o070340.htm','Plateau problem, multi-dimensional','../p/p072850.htm','Quasi-analytic class','../q/q076370.htm')" style="background-color:yellow;">S.N. Bernshtein] . The method was subsequently extended and was systematically utilized in the study of various problems for elliptic and parabolic differential operators [[#References|[3]]], [[#References|[4]]], [[#References|[5]]]. | The method was first utilized by S.N. Bernstein [e theory of functions','../c/c025430.htm','Continuation method (to a parametrized family)','../c/c025520.htm','Euler–Lagrange equation','../e/e036510.htm','Fourier series','../f/f041090.htm','Functions of a real variable, theory of','../f/f042130.htm','Hilbert problems','../h/h120080.htm','Jackson inequality','../j/j054000.htm','Laplace theorem','../l/l057530.htm','Lebesgue constants','../l/l057800.htm','Limit theorems','../l/l058920.htm','Linear elliptic partial differential equation and system','../l/l059180.htm','Lyapunov theorem','../l/l061200.htm','Mathematical statistics','../m/m062710.htm','Minimal surface','../m/m063920.htm','Markov–Bernstein-type inequalities','../m/m110060.htm','Ornstein–Uhlenbeck process','../o/o070240.htm','Orthogonal polynomials','../o/o070340.htm','Plateau problem, multi-dimensional','../p/p072850.htm','Quasi-analytic class','../q/q076370.htm')" style="background-color:yellow;">S.N. Bernshtein] . The method was subsequently extended and was systematically utilized in the study of various problems for elliptic and parabolic differential operators [[#References|[3]]], [[#References|[4]]], [[#References|[5]]]. | ||
Revision as of 10:58, 29 May 2020
method of auxiliary functions
A method which is employed in the theory of linear and non-linear partial differential equations. Bernstein's method consists in introducing certain new (auxiliary) functions, which depend on the solution being sought, and which make it possible to establish a priori estimates of the maximum modulus of the derivatives of this solution of the required order.
A simple example of the application of Bernstein's method is the a priori estimate of the modulus of the derivatives of the solution of the Dirichlet problem for the non-linear (quasi-linear) elliptic equation
$$ \tag{* } \left . \begin{array}{r} { \frac{\partial ^ {2} z }{\partial x ^ {2} } + \frac{\partial ^ {2} z }{\partial y ^ {2} } = f \left ( x, y, z, \frac{\partial z}{\partial x} , \frac{\partial z}{\partial y} \right ) \equiv } \\ {\equiv a \left ( \frac{\partial z}{\partial x} \right ) ^ {2} + 2b \frac{\partial z}{\partial x} \frac{\partial z}{\partial y} + c \left ( \frac{\partial z}{\partial y} \right ) ^ {2} + } \\ {+ 2d \frac{\partial z}{\partial x} + 2e \frac{\partial z}{\partial y} + g \ \mathop{\rm in} \textrm{ the } \textrm{ disc } D, } \\ \end{array} \right \} $$
$$ z \mid _ {C} = 0, $$
where $ a, b, c, d, e, g $ are smooth functions of $ x, y, z $; $ C $ is the circle, the boundary of the disc $ D $ with radius $ R $( the assumption to the effect that $ D $ is a disc and $ z\mid _ {C} = 0 $ is immaterial, since the general case of an arbitrary-connected domain and inhomogeneous boundary condition is readily reduced to the case under consideration by a change of the function and a conformal transformation of the domain).
If $ f _ {z} ^ { \prime } \geq 0 $, then the estimated maximum modulus $ n $
$$ n = \max _ {(x,y) \in D + C } \ | z (x, y) | $$
of the solution of problem (*) is immediately obtained from the maximum principle.
In order to prove that a regular solution of problem (*) exists it is sufficient to have a priori estimates of the maximum modulus of the derivatives of the solution up to the third order (cf. Continuation method (to a parametrized family)). To estimate $ \max _ {C} | \partial z/ \partial x | $ and $ \max _ {C} | \partial z/ \partial y | $, it is sufficient to estimate $ \max _ {C} | \partial z/ \partial \rho | $( since $ z \mid _ {C} = 0 $), where $ \rho , \theta $ are polar coordinates in the disc $ D $. Now introduce a new (auxiliary) function $ u $, given by the formula
$$ z = \phi _ {1} (u) = -n- \alpha + \alpha \mathop{\rm ln} u , $$
where $ \alpha > 0 $ will be selected later. The function $ u = u(x, y) $ varies from $ e $ to $ e ^ {(2n + \alpha )/ \alpha } $ in the same direction as $ z(x, y) $( $ -n \leq z \leq n $). Since
$$ \frac{\partial z}{\partial x} = \frac \alpha {u} \frac{\partial u}{\partial x} , $$
$$ \frac{\partial ^ {2} z }{\partial x ^ {2} } = \ \frac \alpha {u} \frac{\partial ^ {2} u }{\partial x ^ {2} } - \frac \alpha {u ^ {2} } \left ( \frac{\partial u}{\partial x} \right ) ^ {2} , $$
and similar results for derivatives with respect to $ y $, it follows that $ u $ satisfies the equation
$$ \frac{\partial ^ {2} u }{\partial x ^ {2} } + \frac{\partial ^ {2} u }{\partial y ^ {2} } = \ \frac{1}{u} \left [ \left ( \frac{\partial u}{\partial x} \right ) ^ {2} + \left ( \frac{\partial u}{\partial y} \right ) ^ {2} \ \right ] + $$
$$ + \frac \alpha {u} \left [ a \left ( \frac{\partial u}{\partial x} \right ) ^ {2} + 2b \frac{\partial u}{\partial x} \frac{\partial u}{\partial y} + c \left ( \frac{\partial u}{\partial y} \right ) ^ {2} \right ] + $$
$$ + 2d \frac{\partial u}{\partial x} + 2e \frac{\partial u}{\partial y} + g \frac{u} \alpha \equiv Q. $$
Let $ M $ be the upper bound of $ | a |, | b |, | c | $ in $ D $, and let $ \alpha = 1/8 M $. If $ \partial u/ \partial x $ and $ \partial u/ \partial y $ are considered as current coordinates in the plane, and $ x, y, z $ as parameters, the equation $ Q = 0 $ is the equation of an ellipse, since the determinant $ a _ {1} c _ {1} - b _ {1} ^ {2} > 2 u ^ {2} /3 $, where
$$ a _ {1} = \frac{1}{u} \left ( 1+ \frac{a}{8M} \right ) ,\ \ b _ {1} = \frac{b}{8Mu} ,\ \ c _ {1} = \frac{1}{u} \left ( 1 + \frac{t}{8M} \right ) . $$
Thus, for any $ \partial u/ \partial x $ and $ \partial u/ \partial y $, $ Q $ will not be smaller than a certain negative number $ -P $, $ Q \geq -P $( the number $ P $ is readily obtained in explicit form). If one introduces the function $ u _ {1} $ given by the formula
$$ u _ {1} = u + \frac{P}{4} (x ^ {2} +y ^ {2} ), $$
one obtains
$$ \frac{\partial ^ {2} u _ {1} }{\partial x ^ {2} } + \frac{\partial ^ {2} u _ {1} }{\partial y ^ {2} } = Q + P \geq 0, $$
and $ u _ {1} $ attains its maximum on the boundary $ C $ of the domain $ D $ and, since $ u _ {1} $ is constant on $ C $, one has
$$ \frac{\partial u _ {1} }{\partial \rho } \geq 0 \ \ \textrm{ and } \ \ \frac{\partial u}{\partial \rho} \geq \ - \frac{1}{2} PR , $$
where $ R $ is the radius of the circle $ C $. Hence it is possible to find a negative lower bound for $ \partial z/ \partial \rho $:
$$ \frac{\partial z}{\partial \rho} = \ \frac \alpha {u} \frac{\partial u}{\partial \rho} \geq \ - \frac{\alpha PR }{2e ^ {(2n + \alpha )/ \alpha } } . $$
If the same reasoning is applied to a second auxiliary function $ u $
$$ z = \phi _ {2} (u) = \ -n- \alpha + \alpha \mathop{\rm ln} \frac{1}{1-u} , $$
one obtains an estimate from above
$$ \frac{\partial z}{\partial \rho} \leq \ \frac{\alpha P _ {1} R }{2} e ^ {(n+ \alpha )/2 } . $$
Thus, $ \max _ {C} | \partial z/ \partial \rho | $ is estimated, which means that $ \max _ {C} | \partial z/ \partial x | $ and $ \max _ {C} | \partial z/ \partial y | $ are estimated as well. The estimate of the maximum modulus of the first derivatives inside the domain $ D $ is performed in a similar manner: introduce an auxiliary function $ u $ given by the formula
$$ z = \phi _ {3} (u) = -n + \alpha \ \mathop{\rm ln} \mathop{\rm ln} u . $$
The function $ u $ varies in the same direction as $ z $, from $ e $ to $ e ^ {e ^ {2n/ \alpha } } $. In view of (*), on may write the following expression for $ u $
$$ \frac{\partial ^ {2} u }{\partial x ^ {2} } + \frac{\partial ^ {2} u }{\partial y ^ {2} } = \ \frac{1}{u \mathop{\rm ln} u } \left [ (1+ \mathop{\rm ln} u + \alpha a) \left ( \frac{\partial u}{\partial x} \right ) ^ {2} +2ab \frac{\partial u}{\partial x} \frac{\partial u}{\partial y\right} {} + $$
$$ + \left . (1 + \mathop{\rm ln} u + \alpha c) \left ( \frac{\partial u}{\partial y} \right ) ^ {2} \right ] + 2d \frac{\partial u}{\partial x} + 2e \frac{\partial u}{\partial y} + g \frac{u \mathop{\rm ln} u } \alpha \equiv Q _ {1} . $$
Considerations similar to those given above show that if the function
$$ w = \left ( \frac{\partial u}{\partial x} \right ) ^ {2} + \left ( \frac{\partial u}{\partial y} \right ) ^ {2} $$
attains a maximum in the domain $ D $, this maximum does not exceed some number, the value of which depends solely on $ n $ and $ M $. This yields the required estimates of $ \max _ {D} | \partial z/ \partial x | $ and $ \max _ {D} | \partial z/ \partial y | $.
Bernstein's method may also be used to estimate, in a similar manner, the maximum modulus in the domain $ D + C $ of all highest derivatives of the solution (the only other operation which is required is the differentiation of the initial equation).
The method was first utilized by S.N. Bernstein [e theory of functions','../c/c025430.htm','Continuation method (to a parametrized family)','../c/c025520.htm','Euler–Lagrange equation','../e/e036510.htm','Fourier series','../f/f041090.htm','Functions of a real variable, theory of','../f/f042130.htm','Hilbert problems','../h/h120080.htm','Jackson inequality','../j/j054000.htm','Laplace theorem','../l/l057530.htm','Lebesgue constants','../l/l057800.htm','Limit theorems','../l/l058920.htm','Linear elliptic partial differential equation and system','../l/l059180.htm','Lyapunov theorem','../l/l061200.htm','Mathematical statistics','../m/m062710.htm','Minimal surface','../m/m063920.htm','Markov–Bernstein-type inequalities','../m/m110060.htm','Ornstein–Uhlenbeck process','../o/o070240.htm','Orthogonal polynomials','../o/o070340.htm','Plateau problem, multi-dimensional','../p/p072850.htm','Quasi-analytic class','../q/q076370.htm')" style="background-color:yellow;">S.N. Bernshtein] . The method was subsequently extended and was systematically utilized in the study of various problems for elliptic and parabolic differential operators [3], [4], [5].
References
| [1a] | S.N. [S.N. Bernshtein] Bernstein, "Sur la généralisation du problème de Dirichlet (première partie)" Math. Ann. , 62 (1906) pp. 253–271 | 
| [1b] | S.N. [S.N. Bernshtein] Bernstein, "Sur la généralisation du problème de Dirichlet (deuxième partie)" Math. Ann. , 69 (1910) pp. 82–136 | 
| [2] | S.N. Bernshtein, , Collected works , 3 , Moscow (1960) | 
| [3] | O.A. Ladyzhenskaya, N.N. Ural'tseva, "Linear and quasilinear elliptic equations" , Acad. Press (1968) (Translated from Russian) | 
| [4] | A.V. Pogorelov, "Die Verbiegung konvexer Flächen" , Akademie Verlag (1957) (Translated from Russian) | 
| [5] | O.A. Oleinik, S.N. Kruzhkov, "Quasi-linear parabolic equations of second order in several independent variables" Russian Math. Surveys , 16 : 2 (1961) pp. 105–146 Uspekhi Mat. Nauk , 16 : 5 (1961) pp. 115–155 | 
Bernstein method. Encyclopedia of Mathematics. URL: http://encyclopediaofmath.org/index.php?title=Bernstein_method&oldid=13167