Solving linear differential equations is a common problem in almost all fields of science and engineering. Here, we present a variational algorithm with shallow circuits for solving such a problem: given an \(N \times N\) matrix \({\varvec{A}}\) , an N-dimensional vector \(\varvec{b}\) , and an initial vector \(\varvec{x}(0)\) , how to obtain the solution vector \(\varvec{x}(T)\) at time T according to the constraint \(\textrm{d}\varvec{x}(t)/\textrm{d} t = {\varvec{A}}\varvec{x}(t) + \varvec{b}\) . The core idea of the algorithm is to encode the equations into a ground state problem of the Hamiltonian, which is solved via hybrid quantum-classical methods with high fidelities. Compared with the previous works, our algorithm requires the least qubit resources and can restore the entire evolutionary process. In particular, we show its application in simulating the evolution of harmonic oscillators and dynamics of non-Hermitian systems with \(\mathcal{P}\mathcal{T}\) -symmetry. Our algorithm framework provides a key technique for solving so many important problems whose essence is the solution of linear differential equations.
