Selcuk Journal of
Applied Mathematics
Sel¸cuk J. Appl. Math.Vol. 4, No. 2, pp. 5–12, 2003
Boundary rigidity for Riemannian manifolds
Arif AmirovDepartment of Mathematics, Devrek Arts and Sciences Faculty, Karaelmas Uni-versity, Incivez 67100, Zonguldak, Turkey;
e-mail: amirov@karaelmas.edu.tr
Received: October 12, 2003
Summary. In the work question of the uniqueness of the solution of the problem of restoring the Riemannian metric by the distances between the pairs of the points of boundary of the region are inves-tigated. The uniqueness of solution of the problem, up to the diffeo-morphism identical on the boundary of the region, for the sufficiently wide class of the metrics is proven.
Key words: Hodograph, semigeodesic coordinates, inverse kine-matic problem, integral geometry problem, special kinetic equation
2000 Mathematics Subject Classification: 53C65
In this paper the problem of reconstructing a Riemannian metric
g = (gij(x)) in the bounded domain D of space Rn, (n > 1) with the boundary S of class C3 from the distances between the boundary points of the domain in a metric, and connected with it problems are investigated. Here questions connected with the uniqueness of a solution to the problem in question will be discussed. A domain D is called convex with respect to a metric g, if any two points x0, x∈ ¯D,
can be joint by unique geodesic of this metric all points of which with the exception, of possible ends, belong to domain D.
For the points x0, x∈ S let us denote through Hg(x, x0) the dis-tance between the points x0, x in the metric g. Function Hg(x, x0) determined on the set S× S is called the hodograph of the metric g. Problem 1. Determine a metric g in the region D if the hodo-graph Hg(x, x0) is known.
Problem 1 arose in geophysics in connection with the study of distribution of the velocities of propagation of elastic waves inside terrestrial globe and its linearization has, in particular, applications in tomography.
It is easy to show nonuniqueness of a solution of the problem 1. Indeed, let ϕ be the diffeomorphism of region D to itself from class
C1 identical on S. It transforms each simple metric g1 again into a simple metric g2 = ϕ∗g1, in the sense that for any vectors ξ, η∈ TxD,
ξ, η(2)x =ϕ∗ξ, ϕ∗η1ϕ(x) equality holds, where ϕ∗- the differential of the map ϕ,., .ix- scalar product on TxD is determined by the metric
gi, i = 1, 2, D is convex with respect to a metric gi. These two metrics have different families of geodesic, but the same hodograph.
The questions naturally arise:
1) are there other types of the nonuniqueness of solution of prob-lem 1 ?
2) when is a metric determined by its hodograph up to isometry, identical on ?
3) for what classes of metrics hodograph determines a metric uniquely?
Let us clarify the formulation of problem 1 as follows:
Problem 2. Let g1, g2 be two metrics which are convex in D. Does the existence of a diffeomorphism ϕ : D → D follow from the equality H1(x, x0) = H2(x, x0), such that ϕ|S = 1, and g2 = ϕ∗g1, where Hk(x, x0) the hodograph of metric gk, k = 1, 2 and equality
ϕ|S = 1 it means that the mapping ϕ is identical on S?
Positive answer to the question formulated in problem 2, is ob-tained only for few class of metrics (see [1-11]). Below (in theorems 1, 2, 3) it is assumed that the domain D is convex with respect to a metric gk = (gij(k)(x)) ∈ C4( ¯D), k = 1, 2. Furthermore it is assumed that a metrics gk are known on the set Dε/D and coin-cide on Dε/D, where Dε is the neighborhood of set D, ε > 0, i.e
Dε = {x ∈ Rn/d(x, D) < ε}, d(x, D)- the Euclidean distance be-tween the point x∈ Rn and the set D, d(x, D) = inf|x − y|, y ∈ D. Let us note that the last condition is not, generally speaking, re-striction on metrics gk in D if they hodographs coincide. Indeed, it is proved in [ 10 ] that, if H1 and H2 coincide, then in the suitable coordinates g1 and g2 will coincide in the space C2(S). Consequently, it is possible to continue g2 from the boundary S ∈ C3 to Dε/D by equality g2 = g1. Then the metrics g1 and ¯g2 will be from C2(Dε), have the same hodograph and coincide on Dε/D, where ¯g2 = g2when
Theorem 1. Let H1 = H2, then
a) there exists a diffeomorphism ϕ : D → D, that ϕ|S = 1 and
g2= ϕ∗g1, ϕ∈ C3(D),
b) if gk
1i = δ1i, where i = 1, 2, . . . , n; k = 1, 2; δ1i - Kronecker’s
symbols, then metrics g1 and g2 coincide in D.
Let V (x, t) be a solution of the equation
(1) Vtt−
n
i,j=1
aij(x)Vxixj = 0
satisfying the conditions
(2) V, Vt|t=0= 0,
∂V
∂N|∂Dx(0,T )= f (γ, t),
where f continuous on Sx(0, T ), N = N (γ) is outer normal to Sx(0, T ),
γ ∈ S, t ∈ (0, T ), T ≥ T0, T0 -the diameter of the region D in the metric g, (aij(x)) inverse to (gij(x)) matrix.
Let us assume that on C(Sx(0, T )) the operator of reaction is known: Raf = V (γ, t), where a = (aij(x)).
Problem 3. It is necessary to determine vector function (aij(x))n1 from equation (1) in the region D, if the operator Ra is known. Theorem 2. Let
1) the domain D is convex with respect to a metric (gijk(x)), where (gijk(x)) is the inverse of (a(k)ij (x)) matrix
2) the operators of reaction corresponding to matrix- functions ak= (a(k)ij (x)), k = 1, 2 coincide, i.e Ra1 = Ra2. Then
a) there exists a diffeomorphism ϕ : D → D from class C3(D),
such that ϕ|S = 1 and a2 = ϕ∗a1 .
b) if a(k)11 = 1, a(k)1i = 0 , k = 1, 2; i = 2, 3, . . . , n then a(1)ij (x) =
a(2)ij (x) in the region D, i, j = 2, 3, . . . , n.
Problem 1 is connected the following so-called inverse kinematic Problem 4. It is necessary to find a vector function a = (aij(x))n1 in D if, τ (x, x0) is known for points x, x0 ∈ S and satisfies in D the equation (3) n i,j=1 aij(x)τxiτxj = 1,
It is not difficult to see that (3) is the characteristic equation of equation (1), when the corresponding characteristic surface is repre-sented in the form t = τ (x, x0), where τ (x, x0)(see [ 1 ]) is the distance between the points x and x0 in the metric g.
Theorem 3. Let the condition 1) of the theorem 2 be satisfied. Then
if H1 = H2 then
a) there exists a diffeomorphism ϕ : D → D from class C3(D),
such that ϕ|S = 1 and a2 = ϕ∗a1 .
b) under the conditions a(k)11 = 1, a(k)1i = 0 , k = 1, 2; i = 2, 3, . . . , n
matrix - functions a1 and a2 coincide in D.
Assertion b) of theorem 3 is the key result of this paper and there is an improvement of the theorem 1 of work [ 5], namely, in this paper the assertion b) is proved without restriction to a metric g of the form (x∈ ¯D, ξ∈ Rn) −1 2 ∂ ∂x1 n i,j=2 gijξiξj ≥ α0|ξ|2, α0 > 0,ξ = (ξ2, ξ3, . . . , ξn). Let us give the outline of the proof of assertion b) of theorem 3. For this let us introduce the following notations relating to the metric
gk= (gij(k)(x)): Γk(x, x0) - ray connecting the points x0∈ ∂D, x ∈ ¯D, a τk(x, x0) - the distance between these points; γ2(x, ζ) - the ray of metric g2 starting from the point x ∈ D in the direction ζ , where (i = 1, 2, . . . , n) ζ = (ζ1, ζ2, . . . , ζn), ζi= n j=1 a(2)ij (x)pj0, pi0 = (τ1(x, x0) + τ2(x, x0))xi.
Let us assume that under the conditions a(k)11 = 1, a(k)1i = 0 ,
i = 2, 3, . . . , n problem 4 has two solutions ak = (a(k)ij ), k = 1, 2, with the same data, i.e. H1 = H2 Then in equation (3) if we first take aij = a(2)ij , and than aij = a(1)ij , and subtracting the resulting equations from each other and transforming the obtained equation correspondingly, for the functions d(x, x0) = τ2(x, x0)− τ1(x, x0) and
bij = a(2)ij (x)− a(1)ij (x) we have (4) n i,j=1 a(2)ij pi0dxj + n i,j=2 bijτ1xiτ1xj = 0.
Let us note that, since a(k)1j = 0, j = 2, 3, . . . , n, a(k)11 = 1, we have b1j = 0, j = 1, 2, . . . , n. It is easy to see that the expression
n
i,j=1a(2)ij pi0dxj is a derivative of d(x, x0) along γ2(x, ζ). Integrating
equality (4) along the ray γ2(x, ζ) and taking into account that, for the points x, x0 ∈ S, d(x, x0) = 0, we will obtain
(5) 0≡ d(x, x0) = γ2(x,ζ) n i,j=2 bijτ1zi(z, x0)τ1zj(z, x0)dt.
Recalling, that ( see [1] )
vi(1)(z, x0)≡ dz (i) (1) dt = n j=1 a(1)ij (z)τ1zj(z, x0), i = 1, 2, . . . , n, from (5) we have (6) d(x, x0) = γ2(x,ζ) n k,l=2 cklvk(1)(z, x0)v(1)l (z, x0)dt, where ckl = n i,j=2bijg(1)ik gjl(1), x, x0 ∈ S.
Equation (6) is a problem of integral geometry for the matrix-function (ckl) but instead of it we will investigate the inverse prob-lem for the special kinetic equation connected with it. In order to formulate it, we need some notations. Let rays Γ2(x, x0), Γ1(x, x0) leave from the point x∈ D at angles ξ, f(ξ) correspondingly. In view of the condition on a metric gk(x), k = 1, 2 for each fixed x ∈ D functions ξ = ξ(x, x0), η = f (ξ) ∈ C3(S1(2)), are invertible and
x0 = x0(ξ)∈ C3(S1(2)), ξ = f−1(η) where S1(2), is the sphere of radius 1 of the metric g2(x) at point x.
Furthermore Jacobian|∂f∂ξ(ξ)| > 0 and x0 ∈ S. Using these facts it is possible to prove that for each fixed x∈ D the equation
(7) F (ξ)≡ ξ + f(ξ) = ζ
is invertible ξ = F−1(ζ), where ξ ∈ S1(2), ζ = F (ξ) ∈ Sv(2), Sv(2) - the sphere of a radius v(ξ) = (ni,j=1a(2)ij (x)pi0pj0)1/2 of the metric g2(x),
F−1(ζ) in the domain of definition. Moreover using the condition that metric gk(x) is written in semigeodesic coordinates, the existence of a function I(ζ) = (I2(ζ), . . . , In(ζ))∈ C3 such, that
(8) vi(1)(x, x0) = Ii(ζ), i = 2, 3, . . . , n
with
where Rn0 = Rn/{0} is Rn without the origin of coordinates, can be proven. Here and subsequently through ξ(ζ) is denoted the vector ξ∈ Rn (ζ∈ Rn) without the first component. In this case if function
f (ξ) is continued from the set S(2)1 by the formula f (lξ) = lf (ξ) , (l > 0, ξ ∈ S1(2) ) then functions F (ξ), F−1(ζ), I(ζ) will be also
homogenous functions.
Let us introduce a function u(x, ζ) according to the formula
u(x, ζ) = γ2(x,ζ) n i,j=2 cijIi(˙z)Ij(˙z)dt,
where ˙z = (dzdt2,dzdt3, . . . ,dzdtn). Differentiating u(x, ζ) at point x in the direction ζ we have the following kinetic equation
(10) n i=1 ζiuxi+ n i,j,k=1 Γijkζiζjuζk = n i,j=2 cijIi(ζ)Ij(ζ),
where Γijk - Christoffel symbols of the metric g2. It follows from equal-ity (6) and (8) that for any x∈ S we have u(x, ζ) = 0.
Thus for the determination u(x, ζ) and a matrix- function (cij), we have the inverse problem (10), u(x, ζ) = 0. Modifying work technique [5] from relations (10), u(x, ζ) = 0 using the condition (9), it is proven, that cij = 0, i.e. a(1)ij = a(2)ij , i, j = 1, 2, . . . , n. Assertion b) of theorem 3 is proved.
Outline of the proof of theorem 1. Assertion b) of theorem 1 fol-lows from assertion b) of theorem 3. Let us prove now assertion a) of theorem 1. Let problem 2 have two solutions, with the same data i.e.,H1 = H2. In the region D for the metric gk, k = 1, 2, we in-troduce semigeodesic coordinates as follows. Let us select any point
V0∈ Dε/ ¯D and let us consider the geodesics outgoing from it. The so-called geodesic hyperspheres orthogonally intersect these geodesics. We take the ends of the segments of a constant length sk= r on the geodesics, outgoing from V0. These ends form the hypersurface, which is called the geodesic hypersphere of a radius r with center in V0of the metric gk. Let us examine a certain region on the hypersphere (which lies outside ¯D) with the parameters u1k, u2k, . . . , unk−1. We will carry the geodesics to the same parameters, connecting center hyperspheres
V0 with the points of region D. We will characterize position of an arbitrary point L on the geodesic with arc length sk = V0L. Then it is obvious that in view of the condition on the metric gk the vari-ables u1k, u2k, . . . , unk−1, sk form the semigeodesic coordinate system in the region D (see [12]). We will denote subsequently coordinates
u1k, u2k, . . . , unk−1, sk through xk1, xk2, . . . , xkn, or x = (xk1, xk2, . . . , xkn). Let Dk be the domain of semigeodesic coordinates of the metric gk we introduced in the region D, k = 1, 2. It is easy to establish one-to-one correspondence between the regions D1 and D2 as follows: To each point x(1) ∈ D1 one assigns a point x(2) ∈ D2 so that both points become images of the same point in the region D in the semi-geodesic coordinates with respect to corresponding metrics g1 and
g2. Then it is unnecessary to construct the independent coordinate system in each domain D1, D2. Namely, it is possible to transform a coordinate system in the region D2 into the region D1 as follows: To each point x(1)in the region D1, the same coordinates xiare assigned which are already a coordinates in the region D2of the corresponding point x(2).Thus, for the corresponding points x(1) and x(2) we have
x(1)i = x(2)i , i = 1, 2, . . . , n. In this case boundary of the region D1 is mapped into boundary of the region D2. Nevertheless, generally speaking, metrics in both spaces remain different. This means that, from analytical point of view there is one region ˜D in which we have
two different metrics ˜g1, ˜g2. Moreover both metrics are written down in the semigeodesic coordinates in ˜D. It is known that (see [12 ]) ˜gk is written down in the semigeodesic coordinates in ˜D if and only if
˜
gk1i = δ1i, where i = 1, 2, . . . , n; k = 1, 2; δ1i - Kronecker’s symbols. By the condition of theorem 1 on the metrics ˜g1, ˜g2 in the region ˜D.
they have the same hodograph, therefore, by assertion b) of theorem 3 they coincide, i.e. ˜g1 = ˜g2.
Let us build diffeomorphism ϕ according to equalities x(1)i = x(2)i ,
i = 1, 2, . . . , n, as follows: let, us assign to a point x1 ∈ D with the
semigeodesic coordinates (u11, u21, . . . , un1−1, s1) in the metric g1 the point x2 ∈ D with the semigeodesic coordinates (u12, u22, . . . , un2−1, s2) in the metric g2 if the equalities ui2 = ui1, 1, 2, . . . , n, s2 = s1 hold. It is not difficult to see that ϕ|S = 1. Actually, for the point x ∈ S the rays Γ1(x, V0), Γ2(x, V0) outside of the region D coincide, since outside D we have g1 = g2, where Γk(x, V0) - geodesic of the metric
gk connecting points x ∈ D and V0, k = 1, 2. Here, in construction of semigeodesic coordinates for the metrics g1 and g2 we take the same geodesic hypersphere with the center at the point V0 which lies outside D. Then by the uniqueness of the ray Γk(x, V0) and from the definition of the coordinates u11, u21, . . . , un1−1, the first (n− 1)
components of the semigeodesic coordinates of the point x∈ S in the metrics g1and g2coincide, i.e ui2= ui1, 1, 2, . . . , n−1. But the equality of last components (s2 = s1) follows from the equality H1 = H2 and from the fact that rays Γ1(x, V0) and Γ2(x, V0) outside of D coincide. Consequently, we have:
1) for x∈ S, ϕ(x) = x,
2) the regions D1 and D2 are coincide, i.e. D1= D2 = ˜D.
Then taking into account 2), the convexity of domain D with re-spect to gk, k = 1, 2, and determination of ϕ we will obtain, that ϕ transforms D to itself. According to the theorem about continuous-differentiability dependence of the solution to the Cauchy problem (determining the ray of the metric g(k)∈ C(4)(D), k = 1, 2) from the initial data and to condition S ∈ C3 and also from the determination of the ϕ we have, that ϕ∈ C3(D). Equality ˜g1 = ˜g2, the
determina-tion of mappings ϕ and ϕ∗ give us, g2 = ϕ∗g1. This proves Theorem
1.
Assertion a) of Theorem 3 follows from the proof of assertion a) of Theorem 1. Theorem 2 is proven using the properties of the fun-damental solution of equation (1) and the assertion of theorem 1.
For the number of the inverse problems of the theory of scatter-ing and spectral analysis of those connected with problem 2, from theorem 1 it is possible to obtain the corresponding results.
References
1. Romanov, V.G. (1980): Inverse problems of mathematical physics. [in Rus-sian], Nauka, Moscow.
2. Mukhometov, R. G. (1981): Problem of restoring the Riemannian metric, Sib. Mat. J.,22, 119–135.
3. Bernstein, I. N. and Gerver M.L. ( 1980): Of the condition of the discern-ability of metrics on the hodographs, in: Methods and the algorithms of the interpretation of seismological data.M.:Nauka, 50–73.(comp. seismology;Iss. 13)
4. Amirov, A. Kh. (1990): Integral geometry and the problem of restoring the Riemannian metric, Dokl. Akad. Nauk USSR,312, 6, 1289–1291
5. Amirov, A. Kh. (2001): To the problem about the restoration of Riemannian metric, Dokl. Akad. Nauk,312, 6, 1289–1291
6. Sharafutdinov, V. A. (1994): Integral geometry of tensor fields , VSP,
Utrecht, the Netherlands.
7. Croke, C. (1991): Rigidity and the distance between boundary points, J. Differential Geom.,33, No. 2, 445-464.
8. Croke, C., Dairbekov N. S. , Sharafutdinov V . A . (2000): Local boundary rigidity of a compact Riemannian manifold with curvature bounded above, Transactions AMS,352, No. 9, 3937–3956.
9. Gromov, M. (1983): Filling Riemannian manifolds, J. Differential Geom.,18, 1-147.
10. Michel, R. (1981): Sur la ridigite imposee par la longueur des of geodesiques, lnvent. Mat.,65, 71-83.
11. Stefanov P. , Uhlmann G. (1998): Rigidity for metrics with the same lengths of geodesics, Math. Res. Lett.,5, 83-96.
12. Rashevsky, P.K. (1964): Riemannian geometry and tensor analysis . [in