2. Janzen–Rayleigh expansion

Consider an isentropic flow in $d$ dimensions with no external forces of a perfect fluid with a polytopic, ideal gas EoS of adiabatic index $\gamma$,

(2.1)\begin{equation} p \propto {\rho ^{\gamma} } . \end{equation}

The equations governing the flow are the continuity equation,

(2.2)\begin{equation} \frac{{\partial \rho }}{{\partial t}} + \boldsymbol{\nabla} \boldsymbol{\cdot} ( {\rho \boldsymbol{u}} ) = 0 , \end{equation}

and the Euler equation,

(2.3)\begin{equation} \frac{{\partial {\boldsymbol{u}}}}{{\partial t}} + ({\boldsymbol{u}} \boldsymbol{\cdot} \boldsymbol{\nabla}) {\boldsymbol{u}} ={-} \frac{{\boldsymbol{\nabla} p}}{\rho } ={-}\frac{\boldsymbol{\nabla} c^{2}}{\gamma-1}. \end{equation}

Here, $\rho, {\boldsymbol u}$ and $p$ are respectively the mass density, velocity and pressure, $c\equiv (\textrm {d} p/ \textrm {d} \rho )^{1/2}=(\gamma p/\rho )^{1/2}$ is the sound velocity and $\boldsymbol {\nabla }$ is the gradient operator in $d$ dimensions.

We henceforth assume steady flow. Combining (2.1) and (2.2) to eliminate $\rho$ in favour of $c$ then yields

(2.4)\begin{equation} \frac{1}{\gamma-1}{\boldsymbol u} \boldsymbol{\cdot} \boldsymbol{\nabla} {c^{2}} ={-}{c^{2}}\boldsymbol{\nabla} \boldsymbol{\cdot} {\boldsymbol u} . \end{equation}

For our inviscid, steady flow, (2.3) yields the Bernoulli principle, namely, the quantity $w{\boldsymbol {u}}^{2}+c^{2}$ is constant along streamlines, where we define $w\equiv (\gamma -1)/2$.

We consider the flow along a body for a uniform flow incident from infinity,

(2.5a,b)\begin{equation} c( {r \to \infty } ) = {c_{\infty}} \quad \mbox{and}\quad {\boldsymbol{u}}( {r \to \infty } ) = u_{\infty}\hat{\boldsymbol z} , \end{equation}

where $r$ is the radial coordinate, $z$ is the coordinates along the flow, $u=|\boldsymbol {u}|$ is the speed and the subscript $\infty$ indicates the far field region (tending to spatial infinity). Using the boundary conditions (2.5a,b) and the assumption that every streamline starts at infinity, the Bernoulli equation may be written as

(2.6)\begin{equation} w{{\boldsymbol u}^{2}} + c^{2} = w{{\boldsymbol u}_\infty ^{2}} + c_\infty ^{2} . \end{equation}

Considering the subsonic regime, we restrict the discussion to a potential flow, writing the velocity as the gradient of the flow potential $\phi$,

(2.7)\begin{equation} \boldsymbol{u} \equiv \boldsymbol{\nabla} \phi . \end{equation}

Isolating $c^{2}$ from (2.6), substituting it into (2.4) and using the potential (2.7), we obtain a single PDE for the potential $\phi$ (Rayleigh Reference Rayleigh1916),

(2.8)\begin{equation} \tfrac{1}{2}(\boldsymbol{\nabla} \phi) \boldsymbol{\cdot} \boldsymbol{\nabla} {( {\boldsymbol{\nabla} \phi } )^{2}} = [ w {u_\infty ^{2}} + c_\infty ^{2} - w {{{( {\boldsymbol{\nabla} \phi } )}^{2}}} ]{\nabla^{2}}\phi . \end{equation}

We normalize the variables to obtain them in a dimensionless form, by taking $\phi \to (u_{\infty }R)\phi$, and ${\boldsymbol r} \to R{\boldsymbol r}$ (which also normalize the velocity, ${\boldsymbol {u}}\to u_\infty {\boldsymbol {u}}$), where $R$ is a characteristic length scale, for example the radius of a hypersphere. Defining ${\mathcal {M}}_{\infty } \equiv u_{\infty }/c_{\infty }$ as the Mach number at infinity, we then arrive at the dimensionless equation

(2.9)\begin{equation} \tfrac{1}{2}(\boldsymbol{\nabla} \phi) \boldsymbol{\cdot} \boldsymbol{\nabla} {( {\boldsymbol{\nabla} \phi } )^{2}} = [ {w + {{\mathcal{M}}_\infty ^{{-}2}} - w{( {\boldsymbol{\nabla} \phi } )}^{2}}]{\nabla^{2}}\phi . \end{equation}

We supplement this equation for the potential with two boundary conditions (BCs): the uniform incident flow from infinity, and the slip, no-penetration condition on the surface of the body (${\hat {\boldsymbol n}}\boldsymbol {\cdot } {\boldsymbol {u}}=0$), namely

(2.10a,b)\begin{equation} \phi ( {r \to \infty } ) = z = r\cos \theta \quad\mbox{and}\quad \hat{\boldsymbol n}\boldsymbol{\cdot} \boldsymbol{\nabla} \phi = 0 . \end{equation}

Here, $\hat {\boldsymbol n}$ is the normal to the body, and we use hyperspherical coordinates: a radial coordinate $0 \le r \le \infty$, a polar angle $0 \le \theta \le {\rm \pi}$ measured with respect to the uniform flow at infinity and $(d-2)$ additional angles. For axisymmetric flows such as in the case of a hypersphere, the flow is (by symmetry) independent of these additional angles.

To arrive at the JRE, we substitute the expansion

(2.11)\begin{equation} \phi{(\boldsymbol r)} \equiv \sum_{m = 0}^{\infty} {{\phi _m(\boldsymbol r)}{{\mathcal{M}}_{\infty}^{2m}}}, \end{equation}

into (2.9), and isolate the different powers of ${\mathcal {M}}_\infty$. This leads to a recursive set of PDEs for $\phi _m$ (for $m \geq 1$),

(2.12)\begin{align} {\nabla^{2}}{\phi _m} & ={-} w{\nabla^{2}}{\phi _{m - 1}} \nonumber\\ & \quad +w \sum_{\{{m_1},{m_2},{m_3}\} = 0}^{m} {\delta_{{m_1} + {m_2} + {m_3},m - 1} ( {\boldsymbol{\nabla} {\phi _{{m_1}}} \boldsymbol{\cdot} \boldsymbol{\nabla} {\phi _{{m_2}}}} ){\nabla^{2}}{\phi _{{m_3}}}} \nonumber\\ & \quad +\frac{1}{2} \sum_{\{{m_1},{m_2},{m_3}\} = 0}^{m} {\delta_{{m_1} + {m_2} + {m_3},m - 1} \boldsymbol{\nabla} ( {\boldsymbol{\nabla} {\phi _{{m_1}}} \boldsymbol{\cdot} \boldsymbol{\nabla} {\phi _{{m_2}}}} ) \boldsymbol{\cdot} \boldsymbol{\nabla} {\phi _{{m_3}}}}, \end{align}

with an initial equation

(2.13)\begin{equation} {\nabla^{2}}{\phi _0} = 0, \end{equation}

along with BCs

(2.14)\begin{equation} {\phi _m}( {r \to \infty } ) = {\delta _{m,0}} r\cos \theta, \end{equation}

at infinity, and

(2.15)\begin{equation} {\hat{\boldsymbol n}} \boldsymbol{\cdot} \boldsymbol{\nabla} {\phi _m} = 0, \end{equation}

on the body. Here, $\delta _{i,j}$ is the Kronecker delta symbol.

The zeroth-order (2.13) is the Laplace equation for $\phi _0$, corresponding to the incompressible limit ${\mathcal {M}}_\infty \to 0$. For higher orders, (2.12) can be regarded as a set of Poisson equations for $\phi _m$ at each order $m$, with a source term being a function of the lower-order solutions, $\{\phi _i\}^{m-1}_{i=0}$, and their derivatives. In general, the solution to (2.12) at any order $m\geq 1$ under the BCs (2.14) and (2.15) is a sum of an inhomogeneous solution and a homogeneous solution,

(2.16)\begin{equation} \phi_m = \phi_m^{{(in)}} + \phi_m^{{(ho)}}, \end{equation}

where $\phi _m^{{(ho)}}$ solves the Laplace equation.

The JRE (2.11) is constructed by iteratively determining the functions $\phi _m$. One starts by deriving $\phi _0$, obtained as the solution to the zeroth-order Laplace equation (2.13) under the BCs. Increasingly higher-order functions $\phi _m$ are then incrementally derived, by solving the corresponding Poisson equations (2.12) under the same BCs. At each order $m\geq 1$, the inhomogeneous solution $\phi _m^{{(in)}}$ is fixed by the source term in the corresponding (2.12), which is explicitly written once all lower-order functions $\phi _m$ are determined. This solution is then combined with an expansion $\phi _m^{{(ho)}}$ of homogenous solutions, the coefficients of which are determined by applying the BCs to the resulting (2.16).

3. JRE for a hypersphere

Simple solutions exist for the flow around a $d$-dimensional hypersphere, for which the Laplace equation has known analytic solutions, and the Poisson equation is easily solved. Choosing the characteristic length $R$ as the hypersphere radius, the normalized no-penetration BC (2.15) here simplifies to

(3.1)\begin{equation} \partial_r\phi(r=1,\theta) = 0. \end{equation}

Considering the hyperspherical and axial symmetries, it is useful to write the general solution to the Laplace equation (2.13) as an infinite sum of positive and negative radial powers (e.g. Feng, Huang & Yang Reference Feng, Huang and Yang2011)

(3.2)\begin{equation} \sum_{n = 1}^{\infty} {(A_n r^{n}+B_n r^{ - n - d + 2}){\mathcal{J}}_n^{(d)}(\mu)}, \end{equation}

where we define $\mu \equiv \cos \theta$, the normalized Jacobi polynomials

(3.3)\begin{equation} {\mathcal{J}}_n^{(d)}( {\mu } ) \equiv \frac{{J_n^{( {({{d - 3}})/{2},({{d - 3}})/{2}} )}( {\mu } )}}{{J_n^{( {({{d - 3}})/{2},({{d - 3}})/{2}} )}( 1 )}} = \frac{C_n^{({d}/{2}-1)}(\mu)}{{{n+d-3}\choose {n}}} , \end{equation}

and numerical coefficients $A_n$ and $B_n$. Here, $J_n^{(\alpha,\beta )}(\mu )$ and $C_n^{(\alpha )}(\mu )$ are respectively the standard Jacobi and Gegenbauer polynomials.

The source term in (2.12) is a sum of multiple terms, each composed of even derivatives in $\theta$ and odd multiples of functions $\phi$. Therefore, the polar dependence will always exhibit the same symmetry as the zero-order solution $\phi _0$. As the hypersphere is isotropic, the incompressible flow also shows a backward–forward symmetry. We conclude that the functions ${\mathcal {J}}_n^{(d)}$ appearing in $\phi _m$ at all orders $m$ have only odd $n$. Next, we construct the hypersphere JRE by considering increasingly larger orders $m$.

3.1. Zeroth order: $m=0$

For the $m=0$ order, the infinity BC (2.14) allows only the radially linear and decaying terms in the solution (3.2). The no-penetration BC (3.1) restricts the solution further, allowing for only the decaying term proportional to ${\mathcal {J}}_1^{(d)}(\mu ) = \mu$. The solution to the Laplace equation (2.13) around a $d$-dimensional hypersphere thus becomes

(3.4)\begin{equation} \phi _0^{(d)}(r,\theta) = \left[ {r + \frac{1}{{( {d - 1} ){r^{d - 1}}}}} \right]\cos\theta. \end{equation}

Henceforth, we omit the dimension superscript $(d)$ unless necessary.

3.2. First order: $m=1$

The first-order potential, $\phi _1$, is determined by

(3.5)\begin{equation} {\nabla^{2}}{\phi _1} ={-} w{\nabla^{2}}{\phi _{0}} +\tfrac{1}{2} ( \boldsymbol{\nabla} {\phi _{{0}}} ) \boldsymbol{\cdot} \boldsymbol{\nabla} ( \boldsymbol{\nabla} {\phi _{{0}}} )^{2} +w ( \boldsymbol{\nabla} {\phi _{{0}}} )^{2}{\nabla^{2}}{\phi _{0}}, \end{equation}

which, by the zeroth-order (2.13), simplifies to

(3.6)\begin{equation} {\nabla^{2}}{\phi _1} = \tfrac{1}{2} ( \boldsymbol{\nabla} {\phi _{{0}}} ) \boldsymbol{\cdot} \boldsymbol{\nabla} ( \boldsymbol{\nabla} {\phi _{{0}}} )^{2}. \end{equation}

Plugging the $m=0$ solution (3.4) into (3.6) gives the Poisson equation with the explicit source term,

(3.7)\begin{align} {\nabla^{2}}{\phi _1} & = [{-}3\mu\!+\!(2+d)\mu^{3}]r^{{-}d-1}\frac{d }{d-1} + [({-}3+d)\mu\!+\!(2\!+\!d-d^{2})\mu^{3}]r^{{-}2d-1}\frac{2d}{(d-1)^{2}} \nonumber\\ & \quad +[({-}3+2d)\mu+(2+d-3d^{2}+d^{3})\mu^{3}]r^{{-}3d-1}\frac{d}{(d-1)^{3}} . \end{align}

The orthogonal decomposition of Jacobi polynomials (Chaggara & Koepf Reference Chaggara and Koepf2010) then yields

(3.8)\begin{align} {\nabla^{2}}{\phi _1} & = {\mathcal{J}}_3( {\mu } )r^{{-}d-1} d \nonumber\\ &\quad + [{-}2d {\mathcal{J}}_1( {\mu } ) -(1+d) (d-2){\mathcal{J}}_3( {\mu } )] r^{{-}2d-1} \frac{2d }{(d-1) (d+2)} \nonumber\\ &\quad + [d({-}4+3d) {\mathcal{J}}_1( {\mu } ) - (1+d-d^{2})(d-2){\mathcal{J}}_3( {\mu } )] r^{{-}3d-1} \frac{d}{(d-1)^{2}(d+2)} . \end{align}

This result may be compactly written in the form

(3.9)\begin{equation} {\nabla^{2}}\phi_1 = \sum_{k,n} {s^{({in})}_{1,k,n}{r^{k}}{\mathcal{J}}_n( {\mu } )} , \end{equation}

where $s^{({in})}_{m,k,n}$ are the expansion coefficients of the order-$m$ Poisson source term, for radial order $k$ and angular order $n$; these coefficients are directly read off the right-hand side of (3.8). The OSM provides these and other coefficients explicitly, for $d=2$ and $d=3$ (tables 1-3 therein).

As the Poisson equation is linear, it suffices to solve, for arbitrary $k$ and $n$, the equation

(3.10)\begin{equation} {\nabla^{2}}\phi_{k,n} = {r^{k}}{\mathcal{J}}_n( {\mu } ) . \end{equation}

This equation has the particular solution

(3.11)\begin{equation} \phi _{k,n}^{({in})} = \frac{{{r^{k+2}}{\mathcal{J}}_n( {\mu })}}{{(k+2) ( {k + d} ) - n( {n + d - 2} )}} , \end{equation}

provided that $k \notin \{- d - n, n-2\}$. These two exceptional values of $k$ do not occur at order $m=1$, as seen from (3.8), but they may appear at higher orders, as discussed below in § 3.3. We may now expand the inhomogeneous solution as

(3.12)\begin{equation} \phi^{{(in)}}_1 = \sum_{k,n}{s^{({in})}_{1,k,n}\phi^{{(in)}}_{k,n}} , \end{equation}

where the numerical coefficients $s^{({in})}_{1,k,n}$ are determined by equating (3.8) and (3.9).

From (3.8) and (3.11), we see that the largest (i.e. least negative) power of $r$ for $m=1$ is $k_{max}=-d+1<0$, implying that $\phi ^{{(in)}}_1$ includes only radially decaying terms. For higher orders, (2.12) combines the derivatives of multiple $\phi _m$ functions, but the highest radial power $k_{max}$ in the source term remains no larger than $-d-1$. Consequently, the inhomogeneous solution for all $m\geq 1$ orders includes only radially decaying terms. This conclusion, combined with the BCs, indicates that the homogeneous solution may be expanded with only negative powers of $r$ for any $m>0$,

(3.13)\begin{equation} \phi^{{(ho)}}_{m>0} = \sum_{n = 1}^{\infty} {s^{(ho)}_{m,n} {r^{ - n - d + 2}}{\mathcal{J}}_n(\mu)} , \end{equation}

where $s^{(ho)}_{m,n}$ are numerical coefficients, to be determined below.

The full solution for $m=1$ now becomes

(3.14)\begin{equation} \phi_1 = \phi_1^{{(in)}} + \phi_1^{{(ho)}} = \sum_{k,n}{s_{1,k,n} r^{k} {\mathcal{J}}_n(\mu)}, \end{equation}

where we have introduced the combined numerical coefficients

(3.15)\begin{equation} s_{1,k,n}=\frac{s^{({in})}_{1,k-2,n}}{k(k+d-2)-n(n+d-2)}+s^{(ho)}_{1,n}\delta_{k,-n-d+2} . \end{equation}

These coefficients may now be determined from the no-penetration BC (3.1),

(3.16)\begin{equation} \frac{\partial \phi_1}{\partial r}(r=1) = \sum_{k,n}{k s_{1,k,n} {\mathcal{J}}_n(\mu)} = 0, \end{equation}

implying that

(3.17)\begin{equation} s^{(ho)}_{1,n} = \frac{1}{n+d-2} \sum_{k,n}{\frac{k s^{({in})}_{1,k-2,n}}{k(k+d-2)-n(n+d-2)}}. \end{equation}

As the coefficients $s^{({in})}_{1,k,n}$ are known from (3.8) and (3.9), the solution (3.14) is completely specified by (3.15) and (3.17).

3.3. Higher orders: $m\geq 2$

Substituting $\phi _0$ and $\phi _1$ in the Poisson equation (2.12) for the next $m=2$ order results again in an equation of the form (3.9), but now for $\phi _2$ and with different coefficients. Equations of the same form persist for higher orders $m$, as long as the source term in (2.12) is free of non-zero terms $s_{m,k,n}$ which satisfy one of the aforementioned special conditions, $k=- d - n$ or $k=n-2$. Indeed, the source term consists of differential operators of the form $\nabla ^{2}f(r,\mu )$ and $\boldsymbol {\nabla } f(r,\mu )\cdot \boldsymbol {\nabla } g(r,\mu )$, where $f$ and $g$ are constructed from lower-order $\phi$ terms. As long as $f$ and $g$ are sums of terms, each of which is a product of powers of $r$ and Jacobi polynomials in $\mu$, the source term remains of the form (3.9).

For the special cases $k \in \{- d - n, n-2\}$, the solution to (3.10) is no longer given by (3.11), which formally diverges as its denominator vanishes. It is sufficient to consider the former case, $k=-d-n$, as the latter, $k = n -2$, is then obtained by the transformation $n \to - n - d + 2$, under which ${\mathcal {J}}_n^{(d)}$ remains invariant. Here, the inhomogeneous solution to (3.10) becomes

(3.18)\begin{equation} \phi_{k={-}d-n,n} ={-}\frac{{\varGamma [ {2,( {2n + d - 2} )\ln ( r )} ]}}{{{{( {2n + d - 2} )}^{2}}}} {r^{n}} {\mathcal{J}}_n( {\mu } ) , \end{equation}

where $\varGamma (\xi,\eta )$ is the incomplete gamma function, defined by

(3.19)\begin{equation} \varGamma ( {\xi,\eta} ) = \int_\eta^{\infty} {{t^{\xi - 1}}{\textrm{e}^{ - t}}\, \textrm{d} t} . \end{equation}

Once a logarithmic term from (3.18) appears in the inhomogeneous solution for $\phi _m$, as a result of a special $k$ term in the source-term expansion (3.9), a power series in $r$ as in (3.14) is no longer sufficient. Indeed, when $\xi =\ell +1$ and $\eta =\tau \ln (r)$ for integers $\ell$ and $\tau$,

(3.20)\begin{equation} \varGamma [ {\ell + 1,\tau\ln (r)} ] = \frac{{\ell!}}{{{r^{\tau}}}}\sum_{j = 0}^{\ell} {\frac{{{{[\tau\ln (r) ]}^{j}}}}{{j!}}}, \end{equation}

contains logarithmic, and not only polynomial, radial terms. Since $\ln (r)$ cannot be expanded as a power series of $r^{-1}$ valid over the full $1\leq r<\infty$ range, one must then take into account logarithmic terms in the expansion of $\phi _m$ and in the resulting source terms of higher-order functions.

Generalizing the prototypical source term in (3.9) to include a logarithm of $r$ to some power $\ell$, we must therefore also consider the generalized equation

(3.21)\begin{equation} {\nabla^{2}}\phi_{k,n,\ell} = r^{k} {\mathcal{J}}_n( {\mu } ) \ln^{\ell}(r). \end{equation}

As long as $k, d$, and $n$ do not satisfy one of the special cases

(3.22)\begin{equation} k \in \{ - d - n, n-2\} , \end{equation}

the inhomogeneous solution to this equation is

(3.23)\begin{align} \phi_{k,n,\ell} & = \frac{{{\mathcal{J}}_n( {\mu } )}}{{2n + d - 2}} \left({r^{{-}n-d+2}}\frac{{\varGamma [ {\ell + 1, ( {-d - k - n } )\ln (r)} ]}}{{{{( { - d - k - n} )}^{\ell + 1}}}}\right. \nonumber\\ &\quad \left. - {r^{n}}\frac{{\varGamma [ {\ell + 1,( { n - k -2} )\ln (r)} ]}}{{{{({ n - k-2} )}^{\ell + 1}}}} \right) , \end{align}

whereas for $k=-d-n$ we find

(3.24)\begin{equation} \phi _{k={-} d - n,n,\ell} ={-} {r^{n}}{\mathcal{J}}_n( {\mu } ) \frac{{\varGamma[ {\ell + 2,( {2n + d - 2} )\ln (r)} ]}}{{{{(\ell + 1)( {2n + d - 2} )}^{\ell + 2}}}} . \end{equation}

The case $k=n-2$ is again obtained using the transformation $n \to - n - d + 2$,

(3.25)\begin{equation} \phi _{k={-}n-d+2,n,\ell} ={-} {r^{{-}n-d+2}}{\mathcal{J}}_{n} ( {\mu } )\frac{{\varGamma[ {\ell + 2,( {-2n-d+2} )\ln (r)} ]}}{{{{(\ell + 1)( {-2n-d+2} )}^{\ell + 2}}}} . \end{equation}

The overall solution for $\phi _m$ may now be expanded as the finite sum

(3.26)\begin{equation} \phi_m = \sum_{k,n,\ell}{s_{m,k,n,\ell} r^{k}{\mathcal{J}}_n^{(d)}( {\mu } )\ln^{\ell} (r)}, \end{equation}

where the numerical coefficients $s_{m,k,n,\ell }$ are determined using the BCs in analogy with the above $m=2$ discussion. The expansion (3.26) is complete, as the inhomogeneous solution yields only powers of $r$ and $\ln (r)$, and the homogeneous solution yields only powers of $r$, so the resulting higher-order source terms are again of the form (3.21).

One can prove, by induction, the following rules for the $m\geq 1$ indices in (3.26),

(3.27)\begin{equation} 1-d-2md \leq k \leq 1-d, \end{equation}

and

(3.28)\begin{equation} 1\leq n \leq 1+2m. \end{equation}

Logarithmic terms, if they emerge, are limited to indices

(3.29)\begin{equation} 0\leq \ell \leq m, \end{equation}

because combining the particular solutions (3.24) and (3.25) with the property of the incomplete gamma function (3.20) shows that the highest logarithmic power increases at most by one in each JRE order (Imai Reference Imai1942, which considered only 2-D flows). Additional rules for the logarithmic term are discussed in § 5.

4. Semi-analytical and numerical solvers for disk and sphere flows

We explicitly solve (2.12) analytically and using a computer-aided symbolic algebra program (Wolfram Mathematica; Wolfram, Research Inc 2019; a notebook is provided at https://physics.bgu.ac.il/~wallersh/) for the hypersphere in the $d=2$ and $d=3$ cases; namely, we derive the flows around a 2-D disk and around a 3-D sphere. The normalized Jacobi functions reduce to Chebyshev polynomials ${T_n}$ in two dimensions,

(4.1)\begin{equation} {\mathcal{J}}_n^{(2)}( {\cos \theta } ) = {T_n}( {\cos \theta } ) = \cos ( {n\theta } ) , \end{equation}

and to Legendre polynomials ${P_n}$ in three dimensions,

(4.2)\begin{equation} {\mathcal{J}}_n^{(3)}( {\cos \theta } ) = {P_n}( {\cos \theta } ) . \end{equation}

As discussed in § 3, we consider the generalized expansion (2.11) and (3.26) of the flow potential in each case,

(4.3)\begin{equation} \phi^{{disk}} = \left(r+\frac{1}{r}\right) \cos\theta + \sum_{m=1}^{\infty} {\sum_{\ell=0}^{m} {\sum_{n=1}^{2m+1} {\sum_{k={-}1-4m}^{{-}1} {a_{m,k,n,\ell}{{\mathcal{M}}_\infty^{2m}}{r^{k}}\cos ( {n\theta } ){{\ln }^{\ell}}(r)}}}}, \end{equation}

and

(4.4)\begin{equation} \phi^{{sphere}} =\left(r+\frac{1}{2r^{2}}\right) \mu + \sum_{m=1}^{\infty} {\sum_{\ell=0}^{m} {\sum_{n=1}^{2m+1} {\sum_{k={-}2-6m}^{{-}2} {{b_{m,k,n,\ell}}{{\mathcal{M}}_\infty^{2m}}{r^{k}}{P_{n}}(\mu){{\ln }^{\ell}}(r)}}}} , \end{equation}

where $a$ and $b$ are the expansion coefficients in two and three dimensions, respectively, and the summation index $n$ takes only odd values.

We compute these JREs analytically, following the steps outlined in § 3. The $m=0$ term is given by (3.4). For each order $m>0$, we compute the source terms on the right-hand side of (2.12), and decompose them into a sum of terms proportional to $r^{k} {\mathcal {J}}_n( {\mu } ) \ln ^{\ell }(r)$ as in (3.21). The inhomogeneous solution $\phi ^{{(in)}}_m$ is then obtained as the corresponding sum of solutions of the form (3.23)–(3.25), and added to the homogeneous solution $\phi ^{{(ho)}}_m$ of (3.13). The coefficients ${s^{(ho)}_{m,n}}$ in the latter sum are determined from the no-penetration BC, as demonstrated for $m=1$ in (3.17). This process is repeated until the designated order of the JRE is reached.

The JRE results are also compared with the flow obtained from a numerical solution to the nonlinear equation (2.9). We follow Pham, Nore & Brachet (Reference Pham, Nore and Brachet2005) and use a pseudospectral collocation method (Boyd Reference Boyd2001) for a fast numerical convergence. In the pseudospectral method, the solution to a PDE is expanded in terms of basis functions, each being a product of individual (and typically orthogonal) functions for each coordinate. In collocation methods, the PDE is then evaluated and solved at a set of points, usually taken as the roots of the basis functions. This leads to a set of linear equations for the basis function coefficients, which are solved to give an approximate PDE solution. In general for pseudospectral methods, if the solution is smooth (all derivatives of all orders are continuous) then the convergence is exponential in the number $N$ of collocation points (Boyd Reference Boyd2001), $O(\textrm {e}^{-N})$.

Pham et al. (Reference Pham, Nore and Brachet2005) use the Chebyshev–Fourier set of basis functions, with $\cos (n\theta )$ and $\sin (n\theta )$ with integer $n$ as angular basis functions. As discussed in § 3, the analytical solution for $\phi$ is a sum of $\cos (n\theta )$ terms with odd positive $n$, so we expand the flow potential in odd cosine functions. To achieve the same accuracy as Pham et al. (Reference Pham, Nore and Brachet2005), we need only a fourth of the angular basis functions.

For the radial part, we apply an inversion map $\varrho \equiv r^{-1}$ and expand in Chebyshev polynomials $T_k(\varrho )$, to resolve both small and large $r$ behaviour. The pseudospectral expansion thus becomes

(4.5)\begin{equation} {\phi _{{ps}}} = \phi_0 + \sum_{k=0}^{k_{\max}}\sum_{n=0}^{n_{\max}} {{c_{k,n}}{T_k}(\varrho)\cos [ {( {2n + 1} )\theta } ]}. \end{equation}

Here, $k_{max}+1$ and $n_{max}+1$ are the radial and angular resolution, i.e. the number of collocation points, and $c_{k,n}$ are real coefficients. The BCs for $\phi _{ps}$ are guaranteed by offsetting collocation points from the boundaries and picking a potential $\phi _0$ consistent with the BCs; a simple choice is the incompressible solution (3.4). Since the effective computational domain and symmetry are the same in our 2-D and 3-D frameworks, we expand both flows, around a disk and around a sphere, in the same basis functions (4.5). Appendix § A provides more details on the calculations of $c_{k,n}$ and shows the numerical convergence.

6. Example: an improved axial hodograph approximation

The flow in front of a blunt body has important implications in diverse physical applications, and is well approximated in both subsonic and supersonic regimes by expanding the perpendicular gradients $q\equiv \partial _\theta u_\theta |_{\theta =0}$ in terms of the parallel (normalized) velocity $u=|u_r|$ (Keshet & Naor Reference Keshet and Naor2016). In the subsonic regime of this hodograph approximation, $q=q(u)$ is defined in the $0< u<1$ region between the stagnation point and spatial infinity, and is expanded around the stagnation point in the form $q(u) \approx q_0+q_1u+q_2u^{2}+q_3u^{3}$, which we designate as a third-order hodograph approximation, Hodo(3). It can be shown that $q_0$ does not vary much with respect to the incompressible case, $q_1=-1/2$, and $q_2$ is small with respect to $q_3$ (Keshet & Naor Reference Keshet and Naor2016).

Using the JRE for the sphere, here we calculate the coefficients $q_i$ analytically, for given order $m$. The resulting expressions for the coefficients are provided for $1\leq m\leq 4$, as a function ${\mathcal {M}}_\infty$ and $\gamma$, in table 4 in the OSM. For illustration, table 1 provides the numerical values of these coefficients for the critical Mach number with $\gamma =7/5$. The coefficients converge rapidly, reaching three-digit accuracy for $m=4$. The effect of compressibility can be seen to be small, as $q_2$ and $q_3$ are smaller than $q_0$ and $q_1$ by two orders of magnitude. We verify the small deviation of $q_0$ from its incompressible value $3/2$, the precise result $q_1=-1/2$ for all orders, and that $q_2\approx q_3/3$ is small albeit non-negligible.

Table 1.

Coefficients of the first four terms in the Taylor expansion of $q(u)$, for different JRE orders $m$, at the critical Mach number ${\mathcal {M}}_\infty ={\mathcal {M}}_{\infty }\simeq 0.5619$ for the flow of a $\gamma =7/5$ fluid around a sphere. See the OSM, appendix B for analytic expressions for $q$ with arbitrary ${\mathcal {M}}_\infty$ and $\gamma$.

Figure 3 shows the compressible contribution to the radial velocity along the symmetry axis of a sphere, with and without the JRE corrections, for a flow at the critical Mach number. Results are shown for both $\gamma =7/5$ (a) and $\gamma =5/3$ (b). The two panels slightly differ because they pertain to different ${\mathcal {M}}_\infty$ values; for the same ${\mathcal {M}}_\infty$, the axial flow for different $1<\gamma <2$ values are nearly indistinguishable. The corrected hodograph approximation provides a much better fit to the JRE and to the actual flow; an advantage of this approximations is its smooth transition into the supersonic regime (Keshet & Naor Reference Keshet and Naor2016).

Figure 3.

Compressibility contribution to the velocity along the symmetry axis of a sphere near the critical Mach number, for $\gamma =7/5$ ((a) ${\mathcal {M}}_\infty ={\mathcal {M}}_c\simeq 0.5619$) and $\gamma =5/3$ (b) ${\mathcal {M}}_\infty ={\mathcal {M}}_c\simeq 0.5462$). Results shown for a numerically converged pseudospectral calculation (ps, $8\times 8$ resolution, solid black curve), a first-order JRE (JRE(1), dot-dashed red) and second-order (JRE(2), dot-dash-dashed cyan) and hodograph approximations with (Hodo(4), dotted blue) and without (Hodo(3), dashed green) the $q_2$ coefficient.

7. Example: an approximately universal flow in front of spheroids

It has been suggested (Keshet & Naor Reference Keshet and Naor2016) that the flow in front of an axisymmetric body is well approximated by the flow in front of a sphere, rescaled to give the same nose curvature. Consider general prolate or oblate spheroids of the form $(x^{2}+y^{2})\alpha ^{-1}+z^{2}\alpha ^{-2}=1$, chosen to be axisymmetric along the $\hat {z}$ flow axis and with unit curvature at the nose. We shift the $z$ coordinate such that the resulting spheroid overlaps with the unit sphere at the nose, by $z \to z+1-\alpha ^{2}$. The pseudospectral code (details in § A) is modified to solve the flow around such spheroids.

Figure 4 demonstrates the shapes of a sphere ($\alpha =1$) and of four shifted prolate spheroids ($\alpha \in \{2,3,4,5\}$), along with the incompressible flow along the most prolate, $\alpha =5$ of these cases. Figure 5 shows both the incompressible (a,c) and the critical (b,d) flows in front of these bodies. The normalized velocity $u$ (upper panels) varies with $\alpha$ and ${\mathcal {M}}_\infty$ (and slightly with $\gamma$, see figures 5d and 3). However, a nearly universal result is obtained for the scaled velocity

(7.1)\begin{equation} u_r^{{universal}}(r) \equiv u_r^{(\alpha)}(r,\theta=0)/(r^{{-}1}+1)^{\alpha^{{-}1/4}}, \end{equation}

being approximately insensitive to the Mach number, prolate body profile and value of $\gamma$. For oblate spheroids ($\alpha <1$), the velocity does not scale to the universal curve.

Figure 4.

Different prolate spheroids of noses coincident with the unit sphere, shown in the $y=0$ plane, along with streamlines (black arrows) of an incompressible flow around the most prolate of these bodies; see text.

Figure 5.

Incompressible (a,c) and critical (for the sphere: ${\mathcal {M}}_\infty =0.5619$; b,d) flows in front of the different prolate spheroids of figure 4 (legend), showing the normalized velocity $u$ (a,b) and the scaled velocity $u_r/(r^{-1}+1)^{\alpha ^{-1/4}}$ (c,d). Results are based on the pseudospectral code with a converged (better than line width), $(k_{\max },n_{\max })=(8,8)$ resolution, for $\gamma =7/5$. In the bottom row, we also plot the third-order JRE around a sphere (JRE(3), solid blue), for comparison.

8. Summary and discussion

We generalize the JRE of a hypersphere beyond two dimensions, providing an exhaustive solution that supplements previous approaches with additional, usually necessary, logarithmic radial terms. Such a generalization is found to be essential in three dimensions, required for obtaining the correct solution for the sphere even at third ($m=3$, i.e. ${\mathcal {M}}_\infty ^{6}$) order, although it is apparently not needed for the special case of a disk in two dimensions. The resulting, arbitrary-order JRE is a useful tool for studying various problems, as we demonstrate by generalizing and extending previous solutions for the axial flow of a sphere, and by presenting a simple approximate scaling that generalizes the axial flow for prolate spheroids.

We briefly summarize the analysis as follows. The JRE of a potential, compressible flow is derived in general (in § 2) and around a hypersphere of arbitrary dimension $d$ (in § 3). The expansion is based on combining the continuity equation (2.4) and the Bernoulli equation (2.6) into a single equation (2.9) for the flow potential, $\phi$, which depends on the incident Mach number ${\mathcal {M}}_\infty$ far away from the hypersphere. An expansion of $\phi$ in powers of ${\mathcal {M}}_\infty$ results in a set of recursive equations (2.12), with the zero order being the incompressible solution (3.4). The equations for higher orders, $\phi _m$, are derived as a sum of particular Poisson equations (3.21), with an exhaustive solution that is a finite sum (3.26) of terms combining powers of $r$, Jacobi polynomials ${\mathcal {J}}^{(d)}_n(\cos \theta )$ and, in addition, powers of $\ln {(r)}$ that emerge when the Poisson source terms satisfy one of the special conditions (3.22).

Previous JRE calculations were typically carried out with only part of the general solution, each order being considered as a sum of powers of $r$ and Jacobi polynomials alone. This approach works well in two dimensions, where one does not encounter any divergence, thus avoiding logarithmic terms, at least up to order $m=50$. However, this approach severely limits the JRE in three dimensions, as without logarithmic terms, the $m=3$ order diverges. The inclusion of logarithmic terms allows us to calculate terms much higher than available for any previous work in three dimensions (Kaplan Reference Kaplan1940; Tamada Reference Tamada1940; Lighthill Reference Lighthill1960; Fuhs & Fuhs Reference Fuhs and Fuhs1976), and with higher accuracy (Frolov Reference Frolov2003). Furthermore, we are able to compute the JRE to arbitrary order in any dimension, providing interesting insights on the flow. For instance, the JRE of the 4-D hypersphere shows logarithmic terms already at order $m=2$, indicating that the 3-D sphere is not unique in requiring such terms.

After deriving the general solutions to all possible Poisson equations that can emerge in the problem ((3.23), (3.24) and (3.25)), one can directly compute the hypersphere JRE to arbitrary order, dimension and EoS. Low-order terms can be derived manually and compared with previous results (e.g. Imai Reference Imai1938,  Reference Imai1941; Shimasaki Reference Shimasaki1955). The derivation of high-order terms can be assisted by a computer-aided symbolic algebra program (e.g. Van Dyke Reference Van Dyke1984), which we provide as a link to a symbolic algebra program (Wolfram, Research Inc 2019). While various physical phenomena are adequately described by low-order JREs, higher orders are needed to demonstrate the emergence of logarithmic terms in some cases (like $m=3$ in the 3-D sphere), and very high orders are needed for precision tests and near the singular points of the expansion. Calculating the high-order expansion symbolically allows us to study with high accuracy the flow at arbitrary subsonic Mach number and equations of state, which is much more efficient than running an individual numerical simulation for each case. We calculate the JRE analytically for general $\gamma$ to order $m=30$ in two dimensions and $m=18$ in three dimensions. For select values of $\gamma$, we proceed to compute $50$ JRE orders in two dimensions and $30$ orders in three dimensions, allowing us to verify previous numerical results of the high-order JRE (Guttmann & Thompson Reference Guttmann and Thompson1993). Using complex variables (e.g. Imai Reference Imai1942; Barsony-Nagy et al. Reference Barsony-Nagy, Er-El and Yungster1987; Rica Reference Rica2001, in two dimensions) leads to simpler recursive PDEs, but yields the same analytic results and does not show numerical advantages over the real-variable JRE.

The logarithmic term emerging at the $m=3$ order in the 3-D sphere JRE is proportional to $5+7\gamma +2\gamma ^{2}$ (table 3 in the OSM), which does not vanish for any $\gamma >0$, including the special case $\gamma =1$ (i.e. $w=0$). This shows that both nonlinear terms on the right-hand side of (2.12) contribute to the logarithm. We find that the highest order of any logarithmic term in the expansion increases by one every three orders in $m$; when this occurs, only one logarithmic term appears, multiplied by a single power of $r$ and a single Legendre polynomial $\propto {r^{-2m-2}P_{2m+1}(\mu )}$ (see table 3 in the OSM for $m=3$). We speculate that these findings are true for higher orders in the JRE and all physical values of $\gamma$ (conjecture 5.2).

We expect the absence of logarithmic terms in the 2-D disk JRE to persist for all orders $m$ and for all $\gamma$ (conjecture 5.1). To elucidate the absence of these terms, consider how logarithmic terms could have putatively emerged in this JRE. The $m=0$ potential $\phi _0=(r+r^{-1})\cos \theta$ induces the $m=1$ Poisson source $2(r^{-7}-2r^{-5})\cos \theta +2r^{-3}\cos 3\theta$ in (2.12), none of whose components satisfies (3.22), so $\phi _1$ has no logarithms. This source curiously lacks terms such as $r^{-5}\cos 3\theta$, which would have produced a logarithm in $\phi _1$. Such a ($k=-5, n=3$) term is conspicuously absent also from the ($m=2, n=3$) Poisson source $[3r^{-11}-2(2+3\gamma )r^{-9}+18\gamma r^{-7}+19r^{-3}]\cos (3\theta )/6$. Inspecting (2.12), one sees several combinations $(m_i,k_i,n_i)$ of terms among $\phi _{{m_1}}, \phi _{{m_2}}$, and $\phi _{{m_3}}$ in the sum which should have produced, in the $\phi _2$ source, a term

(8.1)\begin{equation} {f\begin{pmatrix} m_1 & k_1 & n_1 \\ m_2 & k_2 & n_2 \\ m_3 & k_3 & n_3 \end{pmatrix}} r^{{-}5}\cos 3\theta , \end{equation}

where $f$ are numerical coefficients. However, the sum of these coefficients vanishes,

(8.2)\begin{align} & f \begin{pmatrix} 1 & -3 & 1 \\ 0 & 1 & 1 \\ 0 & 1 & 1 \end{pmatrix}+ f \begin{pmatrix} 0 & -1 & 1 \\ 0 & 1 & 1 \\ 1 & -1 & 1 \end{pmatrix}+ f \begin{pmatrix} 0 & -1 & 1 \\ 1 & -3 & 3 \\ 0 & 1 & 1 \end{pmatrix}+ f \begin{pmatrix} 0 & 1 & 1 \\ 0 & 1 & 1 \\ 1 & -3 & 1 \end{pmatrix} \nonumber\\ & \quad + f \begin{pmatrix} 0 & 1 & 1 \\ 1 & -1 & 3 \\ 0 & 1 & 1 \end{pmatrix} + f \begin{pmatrix} 0 & -1 & 1 \\ 0 & -1 & 1 \\ 1 & -1 & 3 \end{pmatrix} =\frac{3}{2}-\frac{3}{4}+\frac{3}{2}-\frac{3}{4}-\frac{3}{4}-\frac{3}{4}=0 , \end{align}

giving no logarithms in $\phi _2$. Such a cancellation of terms persists at higher orders $m$.

Logarithmic JRE terms emerge in all 3-D objects we examined, including a paraboloid of revolution $\phi _0 = r\cos \theta +\ln {[r(1+\cos \theta )]}/2$ (Kaplan Reference Kaplan1957), the 3-D Rankine half-body $\phi _0 = r\cos \theta +Q/r$ and the Rankine ovoid. They also appear to emerge at higher dimensions, like the $m\geq 2$ logarithmic terms obtained in four dimensions. In contrast, the absence of logarithmic terms is not a property of 2-D flows in general, nor is it unique to the disk. The 2-D disk itself does show logarithmic terms in its JRE if a finite circulation is included, emerging already at order $m=1$ and persisting at higher orders with indices reaching $\ell _{max}=m$ (Hasimoto & Sibaoka Reference Hasimoto and Sibaoka1941; Heaslet Reference Heaslet1944). Logarithmic terms emerge starting at order $m=1$ also in the JRE of the flow around the 2-D Rankine half-body $\phi _0=r\cos \theta +(\sigma /2{\rm \pi} )\ln r$, obtained (e.g. Anderson Reference Anderson2001) by adding a source term of strength $\sigma >0$ to a uniform flow. These Rankine half-body logarithmic terms persist if one adds a sink on the symmetry axis, except in the special case where the sink has strength $(-\sigma )$: the resulting Rankine oval shows no $\ln r$ terms. More generally, 2-D Rankine bodies (defined by a series of sources and sinks on the symmetry axis) show logarithmic terms already at order $m=1$ unless the sum of all source and sink strengths precisely vanishes. The flows around 2-D compact objects with closed-form algebraic prescriptions (Wallerstein & Keshet, in preparation) appear to resemble the disk in showing no logarithmic JRE terms in the absence of circulation.

These results suggest that the emergence of logarithmic JRE terms is directly related to the topological nature of the flow domain. Indeed, we identify logarithmic JRE terms in all the flow domains that are simply connected, and no logarithmic terms in flow domains that are not simply connected. The example of the 2-D disk is particularly interesting: a circulation term ($\propto \theta$) introduces branch cuts that change the non-simply connected domain into a simply connected Riemann surface. The different analytic properties of flows with and without logarithmic JRE terms – whether or not the difference is topological in its essence – can have physical implications for various applications and questions. One example is the validity of attempts to infer the analytic properties of one type of flow based on studies of the other. For instance, it is unclear if the substantial efforts to analyse the transonic controversy based on the 2-D disk are directly relevant to any 3-D body (Wallerstain & Keshet, in preparation). A topological connection would have important implications for a wide range of systems, ranging from the manifestation of the d'Alambert paradox in inviscid flows around spheres moving in superfluids (e.g. Rica Reference Rica2001; Keeling & Berloff Reference Keeling and Berloff2009) to the use of multi-dimensional potential flows as field duals of N-dimensional spacetimes (Keeler, Manton & Monga Reference Keeler, Manton and Monga2020).

Regardless of logarithmic terms, the angular part the solution includes Jacobi polynomials of odd $n$ for all JRE orders, so the flow is symmetric fore and aft of the hypersphere. Hence, as long as the JRE converges and no additional effects are included, there is no net force on the hypersphere, and the flow is drag free for all dimensions. The d'Alembert paradox thus persists in all dimensions.

Even though we calculate the JRE to high orders, many physical features can be adequately captured with only a few low orders (figure 2), even for sonic flow. For example, we find an accuracy better than ${\sim }1\,\%$ in $\boldsymbol {u}$ for $m=5$. The low JRE orders ($m \leq 5$) show that compressibility effects are more prominent in the vicinity of the hypersphere, and in particular at the equatorial plane (figures 1 and 2). High-order JREs, necessary for some applications and for delicate questions such as the transonic controversy, are now straightforward also for the 3-D sphere and in higher dimensions.

With the ability to calculate the JRE to any desired accuracy, we compare it with other approximation methods, such as the hodograph approximation for the flow on the symmetry axis in front of a sphere, which has important physical implications even in the inviscid regime (e.g. Keshet & Naor Reference Keshet and Naor2016). The hodograph approximation of Keshet & Naor (Reference Keshet and Naor2016) performs better (i.e. agrees better with a JRE or the high-resolution pseudospectral solver) than the JRE of order $m=0$, but worse than the JRE of order $m=1$ for any $\gamma$ (figure 3a and 3b). We use the JRE to improve the hodograph approximation (OSM, appendix B), so that it performs better than JRE of order $m=2$ (but not $m=3$) for any $\gamma$ except far ($r\gtrsim 2$) from the sphere; see figure 3(a). The corrected approximation is simple, readily continued to the supersonic regime (Keshet & Naor Reference Keshet and Naor2016) and now also accurate.

It has been speculated that the flow in front of the sphere can be applied to axisymmetric bodies with a similarly scaled nose curvature (Keshet & Naor Reference Keshet and Naor2016, and references therein), such as prolate spheroids (figure 4). While the flows in front of different spheroids are not identical (figures 5a and 5b), they can be approximately mapped onto each other with a universal scaling, independent of $\gamma$ and ${\mathcal {M}}_\infty$, for flows ranging from the incompressible (figure 5c) to the sonic (figure 5d) regimes. The velocity $u_r^{(\alpha )}$ in front of prolate spheroids with semi-axes $\alpha >0$ and $\alpha ^{1/2}$ can be well approximated by a scaled JRE for a sphere

(8.3)\begin{align} u_r^{(\alpha)} (r,\theta=0 )&\approx (r^{{-}1}+1 )^{\alpha^{{-}1/4}-1} u_r^{(1)} (r,\theta=0 )\nonumber\\ &\approx (r^{{-}1}+1 )^{\alpha^{{-}1/4}-1}\sum_{m=0}^{m_{max}} {\mathcal{M}}_\infty^{2m}\partial_{r}\phi_m (r,\theta=0 ). \end{align}

It is natural to ask if such scalings can be identified using the generalized JRE in front of other bodies and in other dimensions, but this is beyond the scope of the present work.