We prove a local Hölder estimate for any exponent $0<\delta <\frac {1}{2}$
0
<
δ
<
1
2
for solutions of the dynamic programming principle
$ u^{\varepsilon } (x) = \sum \limits _{j=1}^{n} \alpha _{j} \underset {\dim (S)=j}{\inf } \underset {|v|=1}{\underset {v\in S}{\sup }} \frac {u^{\varepsilon } (x + \varepsilon v) + u^{\varepsilon } (x - \varepsilon v)}{2} $
u
ε
(
x
)
=
∑
j
=
1
n
α
j
inf
dim
(
S
)
=
j
sup
v
∈
S
|
v
|
=
1
u
ε
(
x
+
ε
v
)
+
u
ε
(
x
−
ε
v
)
2
with α1,αn > 0 and α2,⋯ ,αn− 1 ≥ 0. The proof is based on a new coupling idea from game theory. As an application, we get the same regularity estimate for viscosity solutions of the PDE
$ \sum \limits _{i=1}^{n} \alpha _{i} \lambda _{i}(D^{2}u)=0, $
∑
i
=
1
n
α
i
λ
i
(
D
2
u
)
=
0
,
where λ1(D2u) ≤⋯ ≤ λn(D2u) are the eigenvalues of the Hessian.