betaFeedback

WorksPhysics and Astronomy

Finding the best basis states for the variation after projection nuclear wave functions

Abstract

The variation after projection (VAP) method is expected to be an efficient way of getting the optimized nuclear wave functions, so that they can be as close as possible to the exact shell model ones. However, we found there are two additional problems that may seriously affect the convergence of the VAP iteration. The first problem is, if a randomly selected projected basis state does not mix with a VAP wave function in the VAP calculation, then it is likely that this basis state will never mix with the VAP wave function even after the VAP iteration converges, which means such selected projected basis state is useless. The other problem is the poor orthonormality among the projected basis states that seriously affect the accuracy of the calculated VAP wave function. In the present work, solutions for these two problems are proposed and some examples are presented to test the validity. It turns out that, with the present solutions, the most important projected basis states can be reliably obtained and the fully optimized VAP wave functions can be accurately and efficiently calculated.

Introduction

Nuclei are complicated quantum many-body systems. According to the quantum mechanics, the wave functions of low-lying nuclear states should be obtained by solving Schrödinger’s equation. Practically, this is done by performing the full shell model calculation in a given model space. However, the configuration space can be easily huge in a large model space, which makes the full shell model calculation almost impossible. This difficulty pushes theorists to develop approximated shell model methods, so that the obtained nuclear wave functions are expected to be as close as possible to the exact shell model ones. So far various approximated shell model methods, such as the shell model truncation [1], stochastic quantum Monte Carlo approaches and their extrapolations [4, 5, 6] [2, 3], projected configuration interaction [7], and the class of variation after projection (VAP) methods [8, 9, 10, 11, 12, 13], have been well developed.

Among those approximated shell model methods, the VAP is an important one, which is also believed to have a good shell model approximation in calculating low-lying states of nuclei [8, 9, 10, 11, 12, 13]. Such approximation can be continuously improved by adding more and more projected basis states into that calculated states. Certainly, the added projected states should be important so that the calculated states can be significantly improved. However, the problem is, if an added projected state is randomly generated and does not mix with a calculated state at the beginning of the VAP iteration, then it is likely that this added projected state will never mix with the state even after the VAP iteration converges. This means such added projected basis state is useless and must be abandoned. In this sense, a useful new projected state should mix with a calculated nuclear state at the beginning of the VAP iteration. Therefore, an effective way of how to get a new useful projected states before performing the VAP iteration is crucial in the VAP calculation.

This problem may be solved by previous methods, such as the quantum Monte Carlo diagonalization (QMCD) [3]. In the QMCD, one selects the best new basis state from the stochastically generated candidates. To ensure the effectiveness of the new candidate, shell model hamiltonian is diagonalized in a subspace spanned by all previously selected basis states and the candidate basis. One can then check the contribution of this candidate basis for lowering the energy eigenvalue being calculated. If such contribution is large enough, then this candidate basis will be added to the group of basis states. Otherwise, one needs to check the importance of the next basis candidate. After that, only important basis states were selected and a good approximation is expected. However, in such basis selection, one might sometimes check a large amount of useless candidates before a new important basis is identified. Completely different from the QMCD, here, we propose a new reliable and efficient algorithm, in which a randomly generated useless projected state can be varied so that it may become a new important basis state for the VAP calculation.

The second problem in the VAP calculation is the poor orthonormality among the projected basis states, which seriously affects the accuracy of the calculated VAP wave functions and consequently damages the stability of the VAP iteration. This problem has been considered in our previous work [13], where the projected states are imposed by two constraints to prevent the appearance of redundant states throughout the VAP iteration. In the present work, we replace those two constraints with a new one, which is expected to be more efficient in keeping the projected basis states in good condition during VAP calculation.

This paper is organized as follows. Section II provides a solution of how to generate the useful projected basis states. Section III discusses the problem of orthonormality among the projected basis states. Section IV shows an example of the present VAP calculations. A brief summary and outlook are presented in Section V.

The important projected basis states

Let’s first address the problem of the important projected basis states in the VAP calculation. We start from the simplified VAP wave function taken from our previous work [13],

∣ΨJπMα(n)(K)⟩=∑i=1nfiJπαP^MKJπ∣Φi⟩,(1)\left|\Psi_{J\pi M\alpha}^{(n)}(K)\right\rangle= \sum_{i=1}^{n} f_{i}^{J\pi\alpha} \hat{P}_{MK}^{J\pi} \left|\Phi_i\right\rangle, \tag*{(1)}

where ∣Φi⟩\left|\Phi_i\right\rangle is a Slater determinant (SD) so that the particle number projection can be omitted here. nn is the number of include ∣Φi⟩\left|\Phi_i\right\rangle SDs. P^MKJπ\hat{P}_{MK}^{J\pi} stands for the product of the angular momentum projection operator, P^MKJ\hat{P}_{MK}^{J}, and the parity projection operator, P^π\hat{P}^{\pi}. The KK number can be randomly chosen in the range of ∣K∣≤J\lvert K\rvert\leq J. α\alpha is used to differ the states with the same JJ, π\pi and MM. The fiJπαf_{i}^{J\pi\alpha} coefficients and the corresponding energy, EJπα(n)E_{J\pi\alpha}^{(n)}, are determined by the following Hill-Wheeler equation,

∑i′=1n(Hii′Jπ−EJπα(n)Nii′Jπ)fi′Jπα=0,(2)\sum_{i'=1}^{n}\left(H_{ii'}^{J\pi}-E_{J\pi\alpha}^{(n)}N_{ii'}^{J\pi}\right)f_{i'}^{J\pi\alpha}=0, \tag*{(2)}

where Hii′Jπ=⟨Φi∣H^P^KKJπ∣Φi′⟩H_{ii'}^{J\pi}=\left\langle\Phi_i\right|\hat{H}\hat{P}_{KK}^{J\pi}\left|\Phi_{i'}\right\rangle and Nii′Jπ=⟨Φi∣P^KKJπ∣Φi′⟩N_{ii'}^{J\pi}=\left\langle\Phi_i\right|\hat{P}_{KK}^{J\pi}\left|\Phi_{i'}\right\rangle. The normalization condition is imposed on the fiJπαf_{i}^{J\pi\alpha} coefficients so that ⟨ΨJπMα(n)(K)|ΨJπMα(n)(K)⟩=1\left\langle\Psi_{J\pi M\alpha}^{(n)}(K)\middle|\Psi_{J\pi M\alpha}^{(n)}(K)\right\rangle=1.

Naturally, one can directly take the trial wave function in Eq. (1) to perform the VAP iteration with the initial ∣Φi⟩\left|\Phi_i\right\rangle SDs randomly generated. As a simple example, we use the trial wave function in Eq. (1) with K=0K=0 to perform the VAP calculations for the ground 0+0^{+} state in 64Ge{}^{64}\mathrm{Ge}. The GXPF1A interaction [14] in the fpfp model space is taken. Thus the parity projection can be omitted in this example. The results are shown in Figure 1. It is seen that in the simplest case of n=1n=1, the VAP iteration converges quite fast and the converged energy is −302.983MeV-302.983\mathrm{MeV}. However, if one takes n=2n=2 with random initial SDs, it is quite possible that the final converged VAP energy is also −302.983MeV-302.983\mathrm{MeV}, which is the same as that with n=1n=1. This strongly implies that one of the two projected SDs does not mix with the converged VAP wave function at all.

VAP iterations for ${}^{64}\mathrm{Ge}$ $0^{+}$

Figure 1. (Color online) The VAP iterations with n=1 (black square), n=2 without mixing (red triangle) and n=2 with mixing (blue circle) for Jπ=0+J^\pi=0^+ in 64Ge{}^{64}\mathrm{Ge}.

Theoretically, if there is a new state, ∣ϕ⟩\left|\phi\right\rangle, that does not mix with a known wave function ∣ψ⟩\left|\psi\right\rangle, then one can easily prove the following identity,

⟨ϕ∣(H^−E)∣ψ⟩=0,(3)\left\langle\phi\right|(\hat{H}-E)\left|\psi\right\rangle=0, \tag*{(3)}

where E=⟨ψ∣H^∣ψ⟩E=\left\langle\psi\right|\hat{H}\left|\psi\right\rangle. Actually, ∣ϕ⟩\left|\phi\right\rangle is usually not orthogonal to ∣ψ⟩\left|\psi\right\rangle. So, ∣ϕ⟩\left|\phi\right\rangle needs to be orthogonalized by performing the Gram-Schmidt orthogonalization, and the

orthogonalized one can be written as

∣ϕ′⟩=∣ϕ⟩−⟨ψ|ϕ⟩∣ψ⟩,(4)\left|\phi'\right\rangle=\left|\phi\right\rangle-\left\langle\psi\middle|\phi\right\rangle\left|\psi\right\rangle, \tag*{(4)}

so that ⟨ϕ′∣ψ⟩=0\left\langle\phi'|\psi\right\rangle=0. If there is no mixing between ∣ϕ⟩\left|\phi\right\rangle and ∣ψ⟩\left|\psi\right\rangle, then one should have

⟨ϕ′∣H^∣ψ⟩=0.(5)\left\langle\phi'\right|\hat{H}\left|\psi\right\rangle=0. \tag*{(5)}

By substituting ∣ϕ′⟩\left|\phi'\right\rangle into Eq. (5), one can immediately get Eq. (3). Therefore, one can simply introduce a real and nonnegative quantity that can be used to evaluate the coupling strength between the normalized ∣ϕ⟩\left|\phi\right\rangle and ∣ψ⟩\left|\psi\right\rangle states,

c=∣⟨ϕ∣(H^−E)∣ψ⟩∣2.(6)c=\left|\left\langle\phi\right|(\hat{H}-E)\left|\psi\right\rangle\right|^{2}. \tag*{(6)}

Certainly, if c>0c>0, then ∣ϕ⟩\left|\phi\right\rangle definitely mixes with ∣ψ⟩\left|\psi\right\rangle. Thus one may need to consider how to find a ∣ϕ⟩\left|\phi\right\rangle so that the cc value can be large enough.

In the present work, we take the ∣ΨJπMα(n)(K)⟩\left|\Psi_{J\pi M\alpha}^{(n)}(K)\right\rangle wave function in Eq. (1) as ∣ψ⟩\left|\psi\right\rangle and the new candidate projected state P^MKJπ∣Φ⟩⟨Φ∣P^KKJπ∣Φ⟩\frac{\hat{P}_{MK}^{J\pi}\left|\Phi\right\rangle}{\sqrt{\left\langle\Phi\right|\hat{P}_{KK}^{J\pi}\left|\Phi\right\rangle}} as ∣ϕ⟩\left|\phi\right\rangle. Therefore, one can calculate the following quantity

cα=∣⟨Φ∣P^KMJπ(H^−EJπα(n))∣ΨJπKα(n)(K)⟩∣2⟨Φ∣P^KKJπ∣Φ⟩,(7)c_{\alpha}=\frac{\left|\left\langle\Phi\right|\hat{P}_{KM}^{J\pi}\left(\hat{H}-E_{J\pi\alpha}^{(n)}\right)\left|\Psi_{J\pi K\alpha}^{(n)}(K)\right\rangle\right|^{2}}{\left\langle\Phi\right|\hat{P}_{KK}^{J\pi}\left|\Phi\right\rangle}, \tag*{(7)}

for each calculated ∣ΨJπMα(n)(K)⟩\left|\Psi_{J\pi M\alpha}^{(n)}(K)\right\rangle state.

More generally, when one calculates mm lowest states simultaneously using the algorithm in Ref. [12], then the candidate projected state is useful as long as one of the cαc_{\alpha} quantities is large enough. Hence one can define a global CC quantity as,

C=∑α=1mcα.(8)C=\sum_{\alpha=1}^{m}c_{\alpha}. \tag*{(8)}

It is easy to understand that if the CC value is large enough, then the candidate projected state should be important for the calculated states.

Now, let us try to solve the problem of the example in Figure 1. Based on the converged VAP wave function with n=1n=1, whose energy is −302.983MeV-302.983\mathrm{MeV}, we then randomly generate a second projected state PMKJπ∣Φ⟩P^{J\pi}_{MK}|\Phi\rangle and combine it with the projected basis in this n=1n=1 VAP wave function to form a n=2n=2 wave function, which explicit form can be written as

∣ΨJM1(2)(K)⟩=f1JπPMKJπ∣Φ1⟩+f2JπPMKJπ∣Φ⟩.(9)|\Psi^{(2)}_{JM_1}(K)\rangle=f^{J\pi}_{1}P^{J\pi}_{MK}|\Phi_1\rangle+f^{J\pi}_{2}P^{J\pi}_{MK}|\Phi\rangle. \tag*{(9)}

This ∣ΨJM1(2)(K)⟩|\Psi^{(2)}_{JM_1}(K)\rangle wave function will be further optimized fully by simultaneously varying both ∣Φ1⟩|\Phi_1\rangle and ∣Φ⟩|\Phi\rangle SDs, so that improved energy minimum lower than −302.983MeV-302.983\mathrm{MeV} is expected.

However, if a candidate projected state is randomly generated in a huge configuration space, it is likely the CC value is extremely tiny. Indeed, in the present case, the calculated CC value between the random PMKJπ∣Φ⟩P^{J\pi}_{MK}|\Phi\rangle and the converged ∣ΨJM1(1)(K)⟩|\Psi^{(1)}_{JM_1}(K)\rangle is as tiny as 10−1010^{-10}, which is almost equals to zero. This leads to the fact that the second term in Eq. (9) can be neglected, which means the f2Jπf^{J\pi}_{2} coefficient can be very tiny. Consequently, the gradient of the energy for the wave function in Eq. (9) should be small enough to terminate the VAP iteration at the very beginning. Therefore, the VAP wave function can not be further improved.

This forces us to find out an algorithm to vary the candidate projected state so that the CC quantity can be maximized. To make such maximization more efficient, it is necessary to calculate the gradient and Hessian of CC with respect to the variational parameters of ∣Φ⟩|\Phi\rangle. Fortunately, the matrix elements required by such gradient and Hessian are available in the present VAP calculations. This makes it very convenient in the maximization of the CC quantity. Since such maximization of the CC quantity is equivalent to the minimization of the −C-C, in the practical calculation, we prefer to take the latter so that our present algorithm of the minimization in the VAP calculation can be directly adopted.

The minimization iteration of −C-C is performed for the second projected SD in the above example. The results are shown in Figure 2. It is interesting that although the CC value is extremely tiny at the beginning, it becomes quickly large enough that the second projected SD can sufficiently mix with the original VAP wave function with n=1n=1. In this sense, we do not need to wait for the converged CC. According to Figure 2, one can simply take the projected SD at the 4-th iteration so that the computational time for getting the useful projected SD can be considerably saved.

Iteration of the $C$ quantity

Figure 2. (Color online) Iteration of the CC quantity in Eq. (8) between the second candidate projected state and the n=1n=1 converged VAP wave function in Figure 1 for the Jπ=0+J^\pi=0^+ ground state in 64Ge^{64}\mathrm{Ge}.

With this useful projected SD PMKJπ∣Φ⟩P^{J\pi}_{MK}|\Phi\rangle, one can then perform the VAP calculation with Eq. (9). The calculated results are also shown in Figure 1. This time, the VAP energy is indeed lowered from −302.983MeV-302.983\mathrm{MeV} to −303.522MeV-303.522\mathrm{MeV}.

With this new method, important projected SDs can be selected one by one so that the VAP iteration can be carried out normally. Thus the VAP wave functions can be continuously improved. At this point, one can naturally imagine that if the VAP process is omitted, then the formed nuclear wave function is very similar to that in the Monte Carlo shell model (MCSM) [?]. However, the major difference is, in the MCSM, the important projected SDs are selected from a large amount of the candidates which are generated stochastically. Therefore, it seems that we have proposed an alternative method of basis selection, which can be used to take place of that in the MCSM.

However, before we put more projected states into the VAP wave functions, we should solve another problem that is associated with the orthonormality among the included projected basis states, as will be addressed in the next section.

Orthonormality among the projected basis states

A natural deficiency of the projected states is the poor orthonormality among themselves. A direct consequence of such deficiency is the possible appearance of the redundant projected states, which seriously affects the stability of VAP iteration. This problem of orthonormality has been addressed in Ref. [13]. In that work, the VAP calculation is performed so that the following QQ quantity can be minimized,

Q=∑α=1mEαJπ+χ1∑i=1n1NiiJπ+χ22∑i,j=1i≠jnNijJπNjiJπNiiJπNjjJπ,(10)Q=\sum_{\alpha=1}^{m}E^{J\pi}_{\alpha}+\chi_1\sum_{i=1}^{n}\frac{1}{N^{J\pi}_{ii}}+\frac{\chi_2}{2}\sum_{\substack{i,j=1\\i\ne j}}^{n}\frac{N^{J\pi}_{ij}N^{J\pi}_{ji}}{N^{J\pi}_{ii}N^{J\pi}_{jj}}, \tag*{(10)}

where NijJπ=⟨Φi∣PKKJπ∣Φj⟩N^{J\pi}_{ij}=\langle\Phi_i|P^{J\pi}_{KK}|\Phi_j\rangle. The first term is the sum of the calculated state energies, the last two terms are the constraints which are expected to keep the projected basis states in a good condition so that redundant states may not appear. Actually, one can easily understand that, in Eq. (10), the second term tends to push the NiiJπN_{ii}^{J\pi} norms to large values, and the third term tends to guide the projected basis states to be orthogonal to one another. Here we propose a more reasonable constraint term that can be used to take place of those two constraint terms in Eq. (10) and the new definition of the QQ quantity can be written as,

Q=∑α=1mEαJπ+χ1∣NJπ∣,(11)Q = \sum_{\alpha=1}^{m} E_{\alpha}^{J\pi} + \chi\frac{1}{\lvert N^{J\pi}\rvert}, \tag*{(11)}

where ∣NJπ∣\lvert N^{J\pi}\rvert is the determinant of the norm matrix in Eq. (2). Notice that this ∣NJπ∣\lvert N^{J\pi}\rvert value equals to the product of all eigenvalues of NJπN^{J\pi}. If there is a redundant basis state, then an eigenvalue of NJπN^{J\pi} must be zero. So the ∣NJπ∣\lvert N^{J\pi}\rvert value also must be zero and the constraint term in Eq. (11) becomes infinite which is impossible. This means the redundant state can be strictly forbidden by the new constraint term in Eq. (11). In contrast, the original constraints in Eq. (10) does not have such strong power in forbidding the appearance of redundant states. For instance, suppose there are three normalized basis states ∣ϕ1⟩\lvert\phi_1\rangle, ∣ϕ2⟩\lvert\phi_2\rangle and ∣ϕ3⟩\lvert\phi_3\rangle, with the following relation ⟨ϕ1∣ϕ2⟩=0\langle\phi_1\vert\phi_2\rangle= 0 and ∣ϕ3⟩=12(∣ϕ1⟩+∣ϕ2⟩)\lvert\phi_3\rangle= \frac{1}{\sqrt{2}}(\lvert\phi_1\rangle+ \lvert\phi_2\rangle). Then ∣ϕ3⟩\lvert\phi_3\rangle is clearly a redundant state. In this case, one can easily get the value of the last term in Eq. (10), and it turns out to be 2χ2\sqrt{2}\chi_2, which seems not large enough to strongly restrict the basis states. Furthermore, such 1∣NJπ∣\frac{1}{\lvert N^{J\pi}\rvert} term tends to increase of the eigenvalues of NJπN^{J\pi}, which also plays the role similar to the second term in Eq. (10). Therefore, it is very nice to take the new constraint to keep the precision of the calculated VAP wave functions.

In the present VAP calculations, the QQ value in Eq. (11) needs to be minimized. Thus it would be better if the gradient and the Hessian matrix of the 1∣NJπ∣\frac{1}{\lvert N^{J\pi}\rvert} term are calculated. Formulations of how to evaluate such quantities have been explicitly presented in the Appendix attached below. Fortunately, all the required matrix elements for the 1∣NJπ∣\frac{1}{\lvert N^{J\pi}\rvert} term are actually available because they are originally prepared for the gradient and the Hessian matrix of the VAP energies.

To show the effect of this new 1∣NJπ∣\frac{1}{\lvert N^{J\pi}\rvert} constraint, we perform the VAP calculations in the sdsd model space. Since this model space is quite small, the dimension of the corresponding configuration space is not so large that the randomly selected projected states can easily mix with one another. This makes it very convenient that one can avoid the complexity of combining this constraint with the first problem addressed in the above section. Therefore, one can randomly generate a group of the projected SDs and directly perform the VAP calculation with or without the 1∣NJπ∣\frac{1}{\lvert N^{J\pi}\rvert} constraint.

We randomly generate 10 projected SDs (n=10n=10) with K=0K=0 to construct the lowest 5 Jπ=8+J^\pi=8^+ states (m=5m=5) in 24Mg{}^{24}\mathrm{Mg}.

in 24Mg{}^{24}\mathrm{Mg}. The USDB interaction [15] is adopted. Then the QQ quantity in Eq. (11) is minimized with χ=0\chi=0 and χ=0.01n=10−20\chi=0.01^{n}=10^{-20}, respectively. The associated quantities as functions of the VAP iteration are shown in Figure 3. It is seen that without any constraint, the lowest eigenvalue of the NJπN^{J\pi} shown in Figure 3(a) tends to be smaller and smaller which is very bad in obtaining precise VAP wave functions. Therefore, the gradient of QQ in Figure 3(b) is not accurately calculated and is very difficult to be smaller than 10−310^{-3} MeV, which is the condition of our VAP convergence. Now, let’s perform the same calculation but with χ=10−20\chi=10^{-20}. This time the eigenvalues of the NJπN^{J\pi} becomes not so small and the VAP iteration converges quite fast. The VAP energies are compared with the exact shell model ones and good approximation still can be achieved even with the new constraint.

Calculated quantities as functions of the VAP iteration

Figure 3. (Color online) Calculated quantities as functions of the VAP iteration for the lowest 5 states (m=5m=5) with Jπ=8+J^\pi=8^+ in 24Mg{}^{24}\mathrm{Mg}. 10 projected SDs are adopted to construct the VAP wave functions. The USDB interaction is adopted. (a) The calculated 10 eigenvalues of NJπN^{J\pi}; (b) The absolute value of the gradient of the QQ quantity; and (c) the calculated lowest five energies. The shell model (SM) energies are also shown for comparison.

IV The example of the 4+4^+ states in 56Ni{}^{56}\mathrm{Ni}

Generally, the above two problems may appear simultaneously in a VAP calculation. Suppose that there are mm lowest states that need to be calculated. The VAP calculation can be performed in two steps. The first step is one can generate n=mn=m projected SDs randomly and construct the simplest VAP wave functions for these mm lowest states. Then these wave functions are varied simultaneously so that the QQ quantity with χ≠0\chi\ne0 in Eq. (11) can be minimized. The second step is to improve the VAP wave functions by adding an useful projected state using the algorithm addressed in Section II. Then one can perform the same VAP iteration as that in the first step but with n=m+1n=m+1. In this way, more projected SDs can be effectively added one by one and the approximations of the calculated mm lowest states can be continuously improved.

As a practical example, we calculate the lowest Jπ=4+J^\pi=4^+ states in 56Ni^{56}\mathrm{Ni}. The GXPF1A interaction [14] in the pfpf shell model space is adopted and K=0K=0 is taken. The constraint parameter χ\chi is assumed to be associated with the number of the included projected states and is simply taken to be χ=10−2n\chi=10^{-2n}. Let’s first calculate the yrast state (m=1m=1). The results are shown in Figure 4. It is shown that the calculated energy is lowered steadily with the projected SDs added one by one. This clearly indicates each of the added projected states is indeed important for the lowest Jπ=4+J^\pi=4^+ state. One can also see that, the energy decrease rapidly at the beginning and then becomes slowly and slowly as the nn number increases. This can be understood that, at each nn, the most important projected state should be obtained because it lowers the energy to the maximum in the VAP algorithm. This means, the next added projected state should be less important than the previous one.

Energy of the lowest yrast state of $J^\pi=4^+$ in $^{56}\mathrm{Ni}$ as a function of $n$

Figure 4. (Color online) The lowest (yrast) energy of Jπ=4+J^\pi=4^+ in 56Ni^{56}\mathrm{Ni} calculated with the present VAP as a function of nn, the number of included projected basis states. The shell model (SM) energy is shown for comparison.

Next, let’s calculate the lowest three 4+4^+ states in 56Ni^{56}\mathrm{Ni}, simultaneously. The calculated results are shown in Figure 5. It is seen that all the calculated energies becomes lower and lower as nn increases, which seems very similar to Figure 4. However, one can also see in Figure 5, the energies are not always significantly improved by adding a new projected state. As we have addressed above, if

The lowest three energies of $J^\pi=4^+$ in $^{56}\mathrm{Ni}$ as a function of $n$

Figure 5. (Color online) Similar to the Fig. 4 but for the lowest three energies of Jπ=4+J^\pi=4^+ in 56Ni^{56}\mathrm{Ni}.

the added new projected state does not mix with one of the state wave functions, then the corresponding energy remains unchanged. From Figure 5, one can understand that the selected new projected state can be important at least for one of the calculated states, but may not always be important for all of them. Nevertheless, It is believed the calculated three energies may be sufficiently close to the exact shell model ones if nn is larger enough.

Summary and outlook

The projected wave functions with good quantum numbers are very nice blocks for the construction of the nuclear wave functions, which can be very close to the exact shell model ones. The full optimized nuclear wave functions expanded in terms of these projected states can be obtained through the VAP calculation. However, when the VAP calculations are performed in a large model space, the selected projected basis states are likely scattered in the extremely huge configuration space. This implies that the selected projected state might be useless for the calculated states. To solve this problem, we propose an algorithm that the useless projected states can be varied so that it can be important for the calculated states. Another problem is, the projected basis states are far from the orthonormality, which seriously damages the stability of the VAP iteration. Here, we solve this problem by attaching a new constraint to the sum of the calculated energies. The proposed solutions for the discussed problems have been supported by the calculated examples.

Certainly, the discussed problems are general ones and are independent of the specific form of the projected states. If the VAP wave functions are formed with the projected Hartree-Fock-Bogoliubov vacuum states, one may still confront with these two problems. In that case, we expect the present solutions would be still valid, which will be further investigated in the future.

Acknowledgments

This work is supported by the National Natural Science Foundation of China under Grant No. 11975314, by the Key Laboratory of Nuclear Data foundation (JCKY20220201C158) and the Continuous Basic Scientific Research Project Nos. WDJ-C-2019-13, BJ20002501.

The Gradient and Hessian matrix of the constrain term

For simplicity, let’s rewrite the constraint term as follows,

Q0=1∣N11⋯N1n⋮⋱⋮Nn1⋯Nnn∣=1∣N∣(12)Q_0 = \frac{1}{\begin{vmatrix} N_{11} & \cdots& N_{1n} \\ \vdots& \ddots& \vdots\\ N_{n1} & \cdots& N_{nn} \end{vmatrix}} = \frac{1}{\lvert N\rvert} \tag*{(12)}

where Nij=⟨Φi∣PKπPKπ∣Φj⟩N_{ij}=\langle\Phi_i|P_{K}^{\pi}P_{K}^{\pi}|\Phi_j\rangle. In the VAP method, one can vary the ∣Φi⟩|\Phi_i\rangle states by applying the Thouless theorem to find the best set of VAP basis states, namely [16]

∣Φi⟩=Ne12∑μνdμνβμ,ν†β0,ν†∣Φ0⟩,(13)|\Phi_i\rangle=\mathcal{N}e^{\frac{1}{2}\sum_{\mu\nu}d_{\mu\nu}\beta_{\mu,\nu}^{\dagger}\beta_{0,\nu}^{\dagger}}|\Phi_0\rangle, \tag*{(13)}

where N\mathcal{N} is the normalization parameter. βμ,ν†β0,ν†\beta_{\mu,\nu}^{\dagger}\beta_{0,\nu}^{\dagger} is generally a quasiparticle pair operator corresponding to the fixed ∣Φ0⟩|\Phi_0\rangle HFB vacuum state, but here it refers to a particle-hole operator and ∣Φ0⟩|\Phi_0\rangle is a Slater determinant. dd is a complex skew matrix. The matrix elements dμνd_{\mu\nu} can be a complex number,

dμν=xμν+iyμν.(14)d_{\mu\nu}=x_{\mu\nu}+iy_{\mu\nu}. \tag*{(14)}

where, xμνx_{\mu\nu} and yμνy_{\mu\nu} are real numbers and are the variational parameters. For convenience, we use xαx_\alpha, xβx_\beta, etc. to stand for these variational parameters.

Then, the gradient of Q0Q_0 can be expressed as

∂Q0∂xα=−1∣N∣2∂∣N∣∂xα=−1∣N∣2∑i=1n∣N11⋯N1n∂Ni1∂xα⋯∂Nin∂xα⋮⋱⋮Nn1⋯Nnn∣=−1∣N∣2∑i,k=1n∂Nik∂xαN{i∣k}‾,(15)\begin{aligned} \frac{\partial Q_0}{\partial x_\alpha}=-\frac{1}{\lvert N\rvert^2}\frac{\partial\lvert N\rvert}{\partial x_\alpha} \\ &=-\frac{1}{\lvert N\rvert^2}\sum_{i=1}^{n}\begin{vmatrix} N_{11} & \cdots& N_{1n} \\ \frac{\partial N_{i1}}{\partial x_\alpha} & \cdots& \frac{\partial N_{in}}{\partial x_\alpha} \\ \vdots& \ddots& \vdots\\ N_{n1} & \cdots& N_{nn} \end{vmatrix} \\ &=-\frac{1}{\lvert N\rvert^2}\sum_{i,k=1}^{n}\frac{\partial N_{ik}}{\partial x_\alpha}\overline{N\{i|k\}}, \tag*{(15)} \end{aligned}

where N{i∣k}‾=(−1)i+jN{i∣k}\overline{N\{i|k\}}=(-1)^{i+j}N\{i|k\} is a cofactor of ∣N∣\lvert N\rvert. The submatrix N{i∣k}N\{i|k\} is obtained by removing the iith row and kkth column from the matrix NN. Calculation of ∂Nik∂xα\frac{\partial N_{ik}}{\partial x_\alpha} can be found in Ref. [11].

Similarly, the Hessian of Q0Q_0 can be written as

∂2Q0∂xα∂xβ=21∣N∣3∂∣N∣∂xα∂∣N∣∂xβ−1∣N∣2∂2∣N∣∂xα∂xβ,(16)\frac{\partial^2 Q_0}{\partial x_\alpha\partial x_\beta}=2\frac{1}{\lvert N\rvert^3}\frac{\partial\lvert N\rvert}{\partial x_\alpha}\frac{\partial\lvert N\rvert}{\partial x_\beta}-\frac{1}{\lvert N\rvert^2}\frac{\partial^2\lvert N\rvert}{\partial x_\alpha\partial x_\beta}, \tag*{(16)}

where ∂2∣N∣∂xα∂xβ\frac{\partial^2\lvert N\rvert}{\partial x_\alpha\partial x_\beta} in the second term of (16) can be calculated with the following equation,

∂2∣N∣∂xα∂xβ=∑i=1n∣N11⋯N1n⋮⋱⋮∂2Ni1∂xα∂xβ⋯∂2Nin∂xα∂xβ⋮⋱⋮Nn1⋯Nnn∣+∑i,j=1i≠jn∣N11⋯N1n∂Ni1∂xα⋯∂Nin∂xα⋮⋱⋮∂Nj1∂xβ⋯∂Njn∂xβ⋮⋱⋮Nn1⋯Nnn∣=∑i,k=1n∂2Nik∂xα∂xβN{i∣k}‾+∑i,j,k,li≠j, k<l∣∂Nik∂xα∂Nil∂xα∂Njk∂xβ∂Njl∂xβ∣N{ij∣kl}.(17)\begin{aligned} \frac{\partial^2\lvert N\rvert}{\partial x_\alpha\partial x_\beta} =\sum_{i=1}^{n} \begin{vmatrix} N_{11} & \cdots& N_{1n}\\ \vdots& \ddots& \vdots\\ \frac{\partial^2N_{i1}}{\partial x_\alpha\partial x_\beta} & \cdots& \frac{\partial^2N_{in}}{\partial x_\alpha\partial x_\beta}\\ \vdots& \ddots& \vdots\\ N_{n1} & \cdots& N_{nn} \end{vmatrix} +\sum_{\substack{i,j=1\\i\ne j}}^{n} \begin{vmatrix} N_{11} & \cdots& N_{1n}\\ \frac{\partial N_{i1}}{\partial x_\alpha} & \cdots& \frac{\partial N_{in}}{\partial x_\alpha}\\ \vdots& \ddots& \vdots\\ \frac{\partial N_{j1}}{\partial x_\beta} & \cdots& \frac{\partial N_{jn}}{\partial x_\beta}\\ \vdots& \ddots& \vdots\\ N_{n1} & \cdots& N_{nn} \end{vmatrix} \\ &=\sum_{i,k=1}^{n}\frac{\partial^2N_{ik}}{\partial x_\alpha\partial x_\beta}\overline{N\{i|k\}} +\sum_{\substack{i,j,k,l\\i\ne j,\ k<l}} \begin{vmatrix} \frac{\partial N_{ik}}{\partial x_\alpha} & \frac{\partial N_{il}}{\partial x_\alpha}\\ \frac{\partial N_{jk}}{\partial x_\beta} & \frac{\partial N_{jl}}{\partial x_\beta} \end{vmatrix} N\{ij|kl\}. \tag*{(17)} \end{aligned}

The ∂2Nik∂xα∂xβ\frac{\partial^2N_{ik}}{\partial x_\alpha\partial x_\beta} also can be found in Ref. [11]. If i<ji<j and k<lk<l, then N{ij∣kl}‾\overline{N\{ij|kl\}} is usually called as the second order cofactor define by

N{ij∣kl}‾=(−1)i+j+k+l∣N{ij∣kl}∣,(18)\overline{N\{ij|kl\}}=(-1)^{i+j+k+l}\lvert N\{ij|kl\}\rvert, \tag*{(18)}

where the submatrix N{ij∣kl}N\{ij|kl\} is obtained from the NN matrix by removing the i,ji,jth rows and k,lthk,lth columns.

Both N{i∣k}‾\overline{N\{i|k\}} and N{ij∣kl}‾\overline{N\{ij|kl\}} can be easily obtained by using Jacobi’s identity in the matrix theory [17],

N{i∣k}‾=Nki−1∣N∣,(19)\overline{N\{i|k\}}=N_{ki}^{-1}\lvert N\rvert, \tag*{(19)}
N{ij∣kl}‾=∣Nki−1Nkj−1Nli−1Nlj−1∣∣N∣.(20)\overline{N\{ij|kl\}}= \begin{vmatrix} N_{ki}^{-1} & N_{kj}^{-1}\\ N_{li}^{-1} & N_{lj}^{-1} \end{vmatrix} \lvert N\rvert. \tag*{(20)}

References

References

  1. [4]Noritaka Shimizu, Takashi Abe, Yusuke Tsunoda, Yutaka Utsuno, Tooru Yoshida1, Takahiro Mizusaki, Michio Honma, and Takaharu Otsuka, Prog. Theor. Exp. Phys. 01A205, (2012).
  2. [5]Noritaka Shimizu, Yutaka Utsuno, Takahiro Mizusaki, Takaharu Otsuka, Takashi Abe, and Michio Honma, Phys. Rev. C 82, 061305(R) (2010).
  3. [6]Noritaka Shimizu, Yusuke Tsunoda, Yutaka Utsuno, and Takaharu Otsuka, Phys. Rev. C 103, 014312 (2021).
  4. [7]Z. C. Gao, M. Horoi, and Y. S. Chen, Phys. Rev. C 79, 014311 (2009).
  5. [8]K. W. Schmid, Prog. Part. Nucl. Phys. 52, 565 (2004).
  6. [9]N. Shimizu, Y. Tsunoda, Y. Utsuno, and T. Otsuka, Phys. Rev. C 103, 014312 (2021).
  7. [10]Z. C. Gao, M. Horoi, and Y. S. Chen, Phys. Rev. C 92, 064310 (2015).
  8. [11]T. Ya, Y. He, Z. C. Gao, J. Q. Wang, and Y. S. Chen, Phys. Rev. C 95, 064307 (2017).
  9. [12]J. Q. Wang, Z. C. Gao, Y. J. Ma, and Y. S. Chen, Phys. Rev. C 98, 021301(R) (2018).
  10. [13]Z. C. Gao, Phys. Lett. B 824, 136795 (2022).
  11. [14]M. Honma, T. Otsuka, B.A. Brown, T. Mizusaki, Eur. Phys. J. A 25(1), 499-502 (2005).
  12. [15]B. A. Brown and W. A. Richter, Phys. Rev. C 74, 034315 (2006).
  13. [16]P. Ring and P. Schuck, The Nuclear Many-Body Problem, (Springer-Verlag, New York, 1980).DOI
  14. [17]Roger A. Horn and Charles R. Johnson, Matrix Analysis, (Cambridge University Press 1994).

Paper details

vlnc:W_nd38psda
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents