Namespaces
Variants
Actions

Difference between revisions of "Bernstein method"

From Encyclopedia of Mathematics
Jump to: navigation, search
(Importing text file)
 
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
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b0157201.png" /></td> <td valign="top" style="width:5%;text-align:right;">(*)</td></tr></table>
+
$$ \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 \}
 +
$$
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b0157202.png" /></td> </tr></table>
+
$$
 +
z \mid  _ {C}  = 0,
 +
$$
  
where <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b0157203.png" /> are smooth functions of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b0157204.png" />; <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b0157205.png" /> is the circle, the boundary of the disc <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b0157206.png" /> with radius <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b0157207.png" /> (the assumption to the effect that <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b0157208.png" /> is a disc and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b0157209.png" /> 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).
+
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 <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572010.png" />, then the estimated maximum modulus <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572011.png" />
+
If $  f _ {z} ^ { \prime } \geq  0 $,  
 +
then the estimated maximum modulus $  n $
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572012.png" /></td> </tr></table>
+
$$
 +
= \max _ {(x,y) \in D + C } \
 +
| z (x, y) |
 +
$$
  
 
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 <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572013.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572014.png" />, it is sufficient to estimate <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572015.png" /> (since <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572016.png" />), where <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572017.png" /> are polar coordinates in the disc <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572018.png" />. Now introduce a new (auxiliary) function <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572019.png" />, given by the formula
+
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
 +
 
 +
$$
 +
= \phi _ {1} (u)  = -n- \alpha + \alpha  \mathop{\rm ln}  u ,
 +
$$
 +
 
 +
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
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572020.png" /></td> </tr></table>
+
$$
  
where <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572021.png" /> will be selected later. The function <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572022.png" /> varies from <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572023.png" /> to <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572024.png" /> in the same direction as <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572025.png" /> (<img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572026.png" />). Since
+
\frac{\partial  ^ {2} u }{\partial  x  ^ {2} }
 +
+
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572027.png" /></td> </tr></table>
+
\frac{\partial  ^ {2} u }{\partial  y  ^ {2} }
 +
  = \
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572028.png" /></td> </tr></table>
+
\frac{1}{u}
  
and similar results for derivatives with respect to <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572029.png" />, it follows that <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572030.png" /> satisfies the equation
+
\left [  \left (
 +
\frac{\partial u}{\partial x}
 +
\right )  ^ {2} +
 +
\left (
 +
\frac{\partial u}{\partial y}
 +
\right )  ^ {2} \
 +
\right ] +
 +
$$
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572031.png" /></td> </tr></table>
+
$$
 +
+
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572032.png" /></td> </tr></table>
+
\frac \alpha {u}
 +
\left [ a \left (
 +
\frac{\partial u}{\partial x}
 +
\right )  ^ {2} + 2b
 +
\frac{\partial u}{\partial x}
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572033.png" /></td> </tr></table>
+
\frac{\partial u}{\partial y}
 +
+ c
 +
\left (
 +
\frac{\partial u}{\partial y}
 +
\right )  ^ {2}  \right ] +
 +
$$
  
Let <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572034.png" /> be the upper bound of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572035.png" /> in <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572036.png" />, and let <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572037.png" />. If <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572038.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572039.png" /> are considered as current coordinates in the plane, and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572040.png" /> as parameters, the equation <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572041.png" /> is the equation of an ellipse, since the determinant <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572042.png" />, where
+
$$
 +
+
 +
2d
 +
\frac{\partial u}{\partial x}
 +
+ 2e
 +
\frac{\partial u}{\partial y}
 +
+ g
 +
\frac{u} \alpha
 +
  \equiv  Q.
 +
$$
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572043.png" /></td> </tr></table>
+
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
  
Thus, for any <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572044.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572045.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572046.png" /> will not be smaller than a certain negative number <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572047.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572048.png" /> (the number <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572049.png" /> is readily obtained in explicit form). If one introduces the function <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572050.png" /> given by the formula
+
$$
 +
a _ {1}  =
 +
\frac{1}{u}
 +
\left ( 1+
 +
\frac{a}{8M}
 +
\right ) ,\ \
 +
b _ {1}  =
 +
\frac{b}{8Mu}
 +
,\ \
 +
c _ {1}  =
 +
\frac{1}{u}
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572051.png" /></td> </tr></table>
+
\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
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572052.png" /></td> </tr></table>
+
$$
  
and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572053.png" /> attains its maximum on the boundary <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572054.png" /> of the domain <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572055.png" /> and, since <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572056.png" /> is constant on <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572057.png" />, one has
+
\frac{\partial  ^ {2} u _ {1} }{\partial  x  ^ {2} }
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572058.png" /></td> </tr></table>
+
+
 +
\frac{\partial  ^ {2} u _ {1} }{\partial y  ^ {2} }
  
where <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572059.png" /> is the radius of the circle <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572060.png" />. Hence it is possible to find a negative lower bound for <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572061.png" />:
+
= Q + P  \geq  0,
 +
$$
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572062.png" /></td> </tr></table>
+
and  $  u _ {1} $
 +
attains its maximum on the boundary  $  C $
 +
of the domain  $  D $
 +
and, since  $  u _ {1} $
 +
is constant on  $  C $,
 +
one has
  
If the same reasoning is applied to a second auxiliary function <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572063.png" />
+
$$
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572064.png" /></td> </tr></table>
+
\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 $
 +
 
 +
$$
 +
= \phi _ {2} (u)  = \
 +
-n- \alpha + \alpha
 +
  \mathop{\rm ln} 
 +
\frac{1}{1-u}
 +
,
 +
$$
  
 
one obtains an estimate from above
 
one obtains an estimate from above
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572065.png" /></td> </tr></table>
+
$$
 +
 
 +
\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
 +
 
 +
$$
 +
= \phi _ {3} (u)  = -n + \alpha \
 +
\mathop{\rm ln}  \mathop{\rm ln}  u .
 +
$$
 +
 
 +
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}
  
Thus, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572066.png" /> is estimated, which means that <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572067.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572068.png" /> are estimated as well. The estimate of the maximum modulus of the first derivatives inside the domain <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572069.png" /> is performed in a similar manner: introduce an auxiliary function <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572070.png" /> given by the formula
+
\frac{\partial u}{\partial y\right}
 +
{} +
 +
$$
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572071.png" /></td> </tr></table>
+
$$
 +
+ \left .
 +
(1 +  \mathop{\rm ln}  u + \alpha c) \left (
 +
\frac{\partial u}{\partial y}
 +
\right )  ^ {2}  \right ] + 2d
  
The function <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572072.png" /> varies in the same direction as <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572073.png" />, from <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572074.png" /> to <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572075.png" />. In view of (*), on may write the following expression for <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572076.png" />
+
\frac{\partial u}{\partial x}
 +
+ 2e
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572077.png" /></td> </tr></table>
+
\frac{\partial u}{\partial y}
 +
+ g
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572078.png" /></td> </tr></table>
+
\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
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572079.png" /></td> </tr></table>
+
$$
 +
= \left (
 +
\frac{\partial u}{\partial x}
 +
\right )  ^ {2} +
 +
\left (
 +
\frac{\partial u}{\partial y}
 +
\right )  ^ {2}
 +
$$
  
attains a maximum in the domain <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572080.png" />, this maximum does not exceed some number, the value of which depends solely on <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572081.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572082.png" />. This yields the required estimates of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572083.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572084.png" />.
+
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 <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015720/b01572085.png" /> of all highest derivatives of the solution (the only other operation which is required is the differentiation of the initial equation).
+
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
How to Cite This Entry:
Bernstein method. Encyclopedia of Mathematics. URL: http://encyclopediaofmath.org/index.php?title=Bernstein_method&oldid=13167
This article was adapted from an original article by I.A. Shishmarev (originator), which appeared in Encyclopedia of Mathematics - ISBN 1402006098. See original article