What's wrong with this time integration?

Hi,
I'm trying to integrate elastodynamic equation using leapfrog algorythm, but my solution with initial velocity is accelerating, instead of "slowing down" due to stifness. Here is my code:

# wave equation in 2D

from dolfin import *
import math
import numpy as np

left =  CompiledSubDomain("near(x[0], side) && on_boundary", side = 0.0)
right = CompiledSubDomain("near(x[0], side) && on_boundary", side = 20.0)

mesh = RectangleMesh(Point(0.0, 0.0), Point(20.0, 5.0), 60, 15)

polyDeg = 1
V = VectorFunctionSpace(mesh, "CG", polyDeg)

rho = 10.0
E, nu = 1.0, 0.3
mu, lmbd = Constant(E/(2*(1 + nu))), Constant(E*nu/((1 + nu)*(1 - 2*nu)))

u = TrialFunction(V)
v = TestFunction(V)

d = u.geometric_dimension()
I = Identity(d)

def L(q):
    return lmbd * tr(q) * I + mu * (q + q.T)

k = inner(L(grad(u)),grad(v))*dx

m = rho * inner(u, v) * dx

f = Expression(('0.0','0.0'))
b = inner(f,v) * dx

T = 10
dt = 0.001
nts = int(math.ceil(T/dt))

u = Function(V)
v = Function(V)

u0 = interpolate(Expression(('0','0')), V)
v0 = interpolate(Expression(('0','0')), V)

u0m = u0.vector().array()
v0m = v0.vector().array()

for initC in range(1700, 1952):
    v0m[initC] = 0.1

um1 = u0m
vm1 = v0m

K = assemble(k)
M = assemble(m)
B = assemble(b)

K = -K.array()
B = -B.array()
M = M.array()
Mi = np.linalg.inv(M)

am1 = np.dot(Mi, np.dot(K, u0m))

for ts in range(1, nts) :
    t = ts * dt

K = assemble(k)
M = assemble(m)
B = assemble(b)

K = -K.array()
B = -B.array()
M = M.array()
Mi = np.linalg.inv(M)

um = um1 + vm1 * dt + 0.5 * am1 * dt*dt

for dirBC in range(0, 40):
    um[dirBC] = 0

u.vector()[:] = um
plot(u, mode = "color", key = "displ", title = str(t))

am = np.dot(Mi, np.dot(K, um))

vm = vm1 + 0.5 * (am1 + am) * dt
print(max(vm))

v.vector()[:] = vm
plot(v, mode = "color", key = "velo", range_max=1.0, range_min=0.0, title = str(t))

um1 = um
am1 = am
vm1 = vm

interactive()