Can this solution space be understood? My question is…
0 reactions
Can this solution space be understood?
My question is concerned with square-integrable functions on \[0,1\]. Say I have a finite number of such functions, denoted by S\_j (j runs over finitely many indices), all known. I also have an unknown function c and known real numbers z\_i (i runs over finitely many indices).
I know the values of ∫ e^(-cz_i) S_j dx for all i and j (over the unit interval), and I want to understand the space of possible candidates for c. My reasoning is that I can decompose e^(-cz_i) = a_i + b_i, where a_i lives in the span of the S_j and b_i lives in the orthogonal complement. It is easy to compute a_i, while b_i is fundamentally unknowable.
Assume for simplicity that i=1,2. Then e^(-cz_1z_2) = (a_1 + b_1)^(z_2) = (a_2 + b_2)^(z_1). This basically says that e^(-cz_1z_2) lives in the intersection of two non-linear spaces: (a_1 + b_1)^(z_2) and (a_2 + b_2)^(z_1) where b_1 and b_2 range over the orthogonal complement of the S_j. Ok, so this basically nails down c to a (transformed version of) this intersection, but is there a way of parametrizing this intersection? Even easier: how to compute a single point in this intersection?
I think one can do the following, but maybe it's overcomplicating things, and maybe does not even work: Pick any b_1 in the orthogonal complement. Now, solve (a_1 + b_1)^(z_2) = (a_2 + b_2)^(z_1) for b_2. If b_2 happens to be in the orthogonal complement also, then we are done (we found one point in the intersection). If not, then project the obtained b_2 onto the orthogonal complement. Now solve the same equation for a new b_1, and keep ping-ponging potentially forever. I have a feeling (more of a hope) that this might converge to a point in the intersection, but I'm clueless how to show this (contraction mapping or something similar?).
Any advice on how to proceed would be greatly appreciated! Even a reference where I can take a look, this is really no my forte....
#science
source
My question is concerned with square-integrable functions on \[0,1\]. Say I have a finite number of such functions, denoted by S\_j (j runs over finitely many indices), all known. I also have an unknown function c and known real numbers z\_i (i runs over finitely many indices).
I know the values of ∫ e^(-cz_i) S_j dx for all i and j (over the unit interval), and I want to understand the space of possible candidates for c. My reasoning is that I can decompose e^(-cz_i) = a_i + b_i, where a_i lives in the span of the S_j and b_i lives in the orthogonal complement. It is easy to compute a_i, while b_i is fundamentally unknowable.
Assume for simplicity that i=1,2. Then e^(-cz_1z_2) = (a_1 + b_1)^(z_2) = (a_2 + b_2)^(z_1). This basically says that e^(-cz_1z_2) lives in the intersection of two non-linear spaces: (a_1 + b_1)^(z_2) and (a_2 + b_2)^(z_1) where b_1 and b_2 range over the orthogonal complement of the S_j. Ok, so this basically nails down c to a (transformed version of) this intersection, but is there a way of parametrizing this intersection? Even easier: how to compute a single point in this intersection?
I think one can do the following, but maybe it's overcomplicating things, and maybe does not even work: Pick any b_1 in the orthogonal complement. Now, solve (a_1 + b_1)^(z_2) = (a_2 + b_2)^(z_1) for b_2. If b_2 happens to be in the orthogonal complement also, then we are done (we found one point in the intersection). If not, then project the obtained b_2 onto the orthogonal complement. Now solve the same equation for a new b_1, and keep ping-ponging potentially forever. I have a feeling (more of a hope) that this might converge to a point in the intersection, but I'm clueless how to show this (contraction mapping or something similar?).
Any advice on how to proceed would be greatly appreciated! Even a reference where I can take a look, this is really no my forte....
#science
source
| 2,000 mlx | |
| 2 | |
| 0 |